identifier
stringlengths 7
18
| space
stringclasses 4
values | uid
stringlengths 1
6
| arch_str
stringlengths 1
32
| input
stringlengths 8.51k
461k
| target_metric
stringclasses 1
value | val_accuracy
float64 0
95.1
| flops
float64 31.1M
14.7B
| params
float64 227k
50M
| metadata
stringlengths 0
1.46k
| metainformation
stringclasses 1
value |
|---|---|---|---|---|---|---|---|---|---|---|
FBNet_2004
|
FBNet
|
2004
|
2004
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_696[FLOAT, 16x3x3x3]
%onnx::Conv_697[FLOAT, 16]
%onnx::Conv_699[FLOAT, 16x16x1x1]
%onnx::Conv_702[FLOAT, 16x1x3x3]
%onnx::Conv_705[FLOAT, 16x16x1x1]
%onnx::Conv_708[FLOAT, 16x16x1x1]
%onnx::Conv_711[FLOAT, 16x1x3x3]
%onnx::Conv_714[FLOAT, 24x16x1x1]
%onnx::Conv_715[FLOAT, 24]
%onnx::Conv_717[FLOAT, 72x24x1x1]
%onnx::Conv_718[FLOAT, 72]
%onnx::Conv_720[FLOAT, 72x1x5x5]
%onnx::Conv_723[FLOAT, 24x72x1x1]
%onnx::Conv_726[FLOAT, 24x24x1x1]
%onnx::Conv_729[FLOAT, 24x1x3x3]
%onnx::Conv_732[FLOAT, 24x24x1x1]
%onnx::Conv_735[FLOAT, 24x12x1x1]
%onnx::Conv_738[FLOAT, 24x1x3x3]
%onnx::Conv_741[FLOAT, 24x12x1x1]
%onnx::Conv_744[FLOAT, 24x12x1x1]
%onnx::Conv_747[FLOAT, 24x1x5x5]
%onnx::Conv_750[FLOAT, 32x12x1x1]
%onnx::Conv_751[FLOAT, 32]
%onnx::Conv_753[FLOAT, 32x16x1x1]
%onnx::Conv_756[FLOAT, 32x1x5x5]
%onnx::Conv_759[FLOAT, 32x16x1x1]
%onnx::Conv_762[FLOAT, 32x32x1x1]
%onnx::Conv_765[FLOAT, 32x1x5x5]
%onnx::Conv_768[FLOAT, 32x32x1x1]
%onnx::Conv_771[FLOAT, 192x32x1x1]
%onnx::Conv_772[FLOAT, 192]
%onnx::Conv_774[FLOAT, 192x1x5x5]
%onnx::Conv_777[FLOAT, 32x192x1x1]
%onnx::Conv_780[FLOAT, 192x32x1x1]
%onnx::Conv_783[FLOAT, 192x1x3x3]
%onnx::Conv_786[FLOAT, 64x192x1x1]
%onnx::Conv_787[FLOAT, 64]
%onnx::Conv_789[FLOAT, 64x32x1x1]
%onnx::Conv_792[FLOAT, 64x1x5x5]
%onnx::Conv_795[FLOAT, 64x32x1x1]
%onnx::Conv_798[FLOAT, 64x64x1x1]
%onnx::Conv_801[FLOAT, 64x1x5x5]
%onnx::Conv_804[FLOAT, 64x64x1x1]
%onnx::Conv_807[FLOAT, 384x64x1x1]
%onnx::Conv_808[FLOAT, 384]
%onnx::Conv_810[FLOAT, 384x1x5x5]
%onnx::Conv_813[FLOAT, 64x384x1x1]
%onnx::Conv_816[FLOAT, 192x64x1x1]
%onnx::Conv_819[FLOAT, 192x1x3x3]
%onnx::Conv_822[FLOAT, 112x192x1x1]
%onnx::Conv_823[FLOAT, 112]
%onnx::Conv_825[FLOAT, 112x112x1x1]
%onnx::Conv_828[FLOAT, 112x1x5x5]
%onnx::Conv_831[FLOAT, 112x112x1x1]
%onnx::Conv_834[FLOAT, 112x56x1x1]
%onnx::Conv_837[FLOAT, 112x1x3x3]
%onnx::Conv_840[FLOAT, 112x56x1x1]
%onnx::Conv_843[FLOAT, 336x112x1x1]
%onnx::Conv_844[FLOAT, 336]
%onnx::Conv_846[FLOAT, 336x1x3x3]
%onnx::Conv_849[FLOAT, 112x336x1x1]
%onnx::Conv_852[FLOAT, 112x56x1x1]
%onnx::Conv_855[FLOAT, 112x1x3x3]
%onnx::Conv_858[FLOAT, 184x56x1x1]
%onnx::Conv_859[FLOAT, 184]
%onnx::Conv_861[FLOAT, 184x92x1x1]
%onnx::Conv_864[FLOAT, 184x1x5x5]
%onnx::Conv_867[FLOAT, 184x92x1x1]
%onnx::Conv_870[FLOAT, 552x184x1x1]
%onnx::Conv_871[FLOAT, 552]
%onnx::Conv_873[FLOAT, 552x1x5x5]
%onnx::Conv_876[FLOAT, 352x552x1x1]
%onnx::Conv_877[FLOAT, 352]
%onnx::Conv_879[FLOAT, 1504x352x1x1]
%onnx::Conv_880[FLOAT, 1504]
) {
%onnx::Conv_874 = Identity(%onnx::Conv_871)
%onnx::Conv_868 = Identity(%onnx::Conv_859)
%onnx::Conv_865 = Identity(%onnx::Conv_859)
%onnx::Conv_862 = Identity(%onnx::Conv_859)
%onnx::Conv_856 = Identity(%onnx::Conv_823)
%onnx::Conv_853 = Identity(%onnx::Conv_823)
%onnx::Conv_850 = Identity(%onnx::Conv_823)
%onnx::Conv_847 = Identity(%onnx::Conv_844)
%onnx::Conv_841 = Identity(%onnx::Conv_823)
%onnx::Conv_838 = Identity(%onnx::Conv_823)
%onnx::Conv_835 = Identity(%onnx::Conv_823)
%onnx::Conv_832 = Identity(%onnx::Conv_823)
%onnx::Conv_829 = Identity(%onnx::Conv_823)
%onnx::Conv_826 = Identity(%onnx::Conv_823)
%onnx::Conv_820 = Identity(%onnx::Conv_772)
%onnx::Conv_817 = Identity(%onnx::Conv_772)
%onnx::Conv_814 = Identity(%onnx::Conv_787)
%onnx::Conv_811 = Identity(%onnx::Conv_808)
%onnx::Conv_805 = Identity(%onnx::Conv_787)
%onnx::Conv_802 = Identity(%onnx::Conv_787)
%onnx::Conv_799 = Identity(%onnx::Conv_787)
%onnx::Conv_796 = Identity(%onnx::Conv_787)
%onnx::Conv_793 = Identity(%onnx::Conv_787)
%onnx::Conv_790 = Identity(%onnx::Conv_787)
%onnx::Conv_784 = Identity(%onnx::Conv_772)
%onnx::Conv_781 = Identity(%onnx::Conv_772)
%onnx::Conv_778 = Identity(%onnx::Conv_751)
%onnx::Conv_775 = Identity(%onnx::Conv_772)
%onnx::Conv_769 = Identity(%onnx::Conv_751)
%onnx::Conv_766 = Identity(%onnx::Conv_751)
%onnx::Conv_763 = Identity(%onnx::Conv_751)
%onnx::Conv_760 = Identity(%onnx::Conv_751)
%onnx::Conv_757 = Identity(%onnx::Conv_751)
%onnx::Conv_754 = Identity(%onnx::Conv_751)
%onnx::Conv_748 = Identity(%onnx::Conv_715)
%onnx::Conv_745 = Identity(%onnx::Conv_715)
%onnx::Conv_742 = Identity(%onnx::Conv_715)
%onnx::Conv_739 = Identity(%onnx::Conv_715)
%onnx::Conv_736 = Identity(%onnx::Conv_715)
%onnx::Conv_733 = Identity(%onnx::Conv_715)
%onnx::Conv_730 = Identity(%onnx::Conv_715)
%onnx::Conv_727 = Identity(%onnx::Conv_715)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%onnx::Conv_712 = Identity(%onnx::Conv_697)
%onnx::Conv_709 = Identity(%onnx::Conv_697)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_697)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_696, %onnx::Conv_697)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%694 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %694
}
|
val_accuracy
| 0 | 49,904,000 | 1,344,900 |
{'zcp_synflow': 72.66688775063213, 'zcp_zen': 61.702877044677734, 'zcp_epe_nas': 19.87381438510896, 'zcp_fisher': 0.12143491953611374, 'zcp_flops': 49904000.0, 'zcp_grad_norm': 18.908231735229492, 'zcp_grasp': 0.05847454071044922, 'zcp_jacov': -16.073882718993897, 'zcp_l2_norm': 515.8079223632812, 'zcp_nwot': 207.49934907918868, 'zcp_params': 1344900.0, 'zcp_plain': 0.0053270249627530575, 'zcp_snip': 33.82162094116211, 'lat_1080ti_1': 0.6120366537760428, 'lat_1080ti_32': 0.5305349290573519, 'lat_1080ti_64': 0.2975655726850402, 'lat_2080ti_1': 0.6488517928220441, 'lat_2080ti_32': 0.5253876227502335, 'lat_2080ti_64': 0.34662959830342804, 'lat_essential_ph_1': 0.1320754716981132, 'lat_eyeriss': 0.21964681935708644, 'lat_fpga': 0.22169227876784842, 'lat_gold_6226': 0.17841640191329203, 'lat_gold_6240': 0.3519762070407285, 'lat_pixel2': 0.17391304347826086, 'lat_pixel3': 0.25692583607943464, 'lat_raspi4': 0.2985937161888225, 'lat_samsung_a50': 0.09473684210526316, 'lat_samsung_s7': 0.47244094488188976, 'lat_silver_4114': 0.3866855714066619, 'lat_silver_4210r': 0.4172714179016779, 'lat_titan_rtx_1': 0.6200482457668498, 'lat_titan_rtx_32': 0.5393746646739299, 'lat_titan_rtx_64': 0.3993801147705541, 'lat_titanx_1': 0.3219186029615212, 'lat_titanx_32': 0.4453865463217168, 'lat_titanx_64': 0.3024383404796916, 'lat_titanxp_1': 0.598080272974068, 'lat_titanxp_32': 0.4853485443541083, 'lat_titanxp_64': 0.34124446404645864}
| |
FBNet_2218
|
FBNet
|
2218
|
2218
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_605[FLOAT, 16x3x3x3]
%onnx::Conv_606[FLOAT, 16]
%onnx::Conv_608[FLOAT, 16x16x1x1]
%onnx::Conv_611[FLOAT, 16x1x5x5]
%onnx::Conv_614[FLOAT, 16x16x1x1]
%onnx::Conv_617[FLOAT, 16x16x1x1]
%onnx::Conv_620[FLOAT, 16x1x3x3]
%onnx::Conv_623[FLOAT, 24x16x1x1]
%onnx::Conv_624[FLOAT, 24]
%onnx::Conv_626[FLOAT, 24x12x1x1]
%onnx::Conv_629[FLOAT, 24x1x5x5]
%onnx::Conv_632[FLOAT, 24x12x1x1]
%onnx::Conv_635[FLOAT, 72x24x1x1]
%onnx::Conv_636[FLOAT, 72]
%onnx::Conv_638[FLOAT, 72x1x5x5]
%onnx::Conv_641[FLOAT, 24x72x1x1]
%onnx::Conv_644[FLOAT, 144x24x1x1]
%onnx::Conv_645[FLOAT, 144]
%onnx::Conv_647[FLOAT, 144x1x5x5]
%onnx::Conv_650[FLOAT, 24x144x1x1]
%onnx::Conv_653[FLOAT, 24x24x1x1]
%onnx::Conv_656[FLOAT, 24x1x5x5]
%onnx::Conv_659[FLOAT, 32x24x1x1]
%onnx::Conv_660[FLOAT, 32]
%onnx::Conv_662[FLOAT, 192x32x1x1]
%onnx::Conv_663[FLOAT, 192]
%onnx::Conv_665[FLOAT, 192x1x5x5]
%onnx::Conv_668[FLOAT, 32x192x1x1]
%onnx::Conv_671[FLOAT, 192x32x1x1]
%onnx::Conv_674[FLOAT, 192x1x5x5]
%onnx::Conv_677[FLOAT, 64x192x1x1]
%onnx::Conv_678[FLOAT, 64]
%onnx::Conv_680[FLOAT, 384x64x1x1]
%onnx::Conv_681[FLOAT, 384]
%onnx::Conv_683[FLOAT, 384x1x5x5]
%onnx::Conv_686[FLOAT, 64x384x1x1]
%onnx::Conv_689[FLOAT, 64x32x1x1]
%onnx::Conv_692[FLOAT, 64x1x5x5]
%onnx::Conv_695[FLOAT, 64x32x1x1]
%onnx::Conv_698[FLOAT, 192x64x1x1]
%onnx::Conv_701[FLOAT, 192x1x3x3]
%onnx::Conv_704[FLOAT, 64x192x1x1]
%onnx::Conv_707[FLOAT, 64x32x1x1]
%onnx::Conv_710[FLOAT, 64x1x3x3]
%onnx::Conv_713[FLOAT, 112x32x1x1]
%onnx::Conv_714[FLOAT, 112]
%onnx::Conv_716[FLOAT, 112x56x1x1]
%onnx::Conv_719[FLOAT, 112x1x3x3]
%onnx::Conv_722[FLOAT, 112x56x1x1]
%onnx::Conv_725[FLOAT, 672x112x1x1]
%onnx::Conv_726[FLOAT, 672]
%onnx::Conv_728[FLOAT, 672x1x3x3]
%onnx::Conv_731[FLOAT, 184x672x1x1]
%onnx::Conv_732[FLOAT, 184]
%onnx::Conv_734[FLOAT, 184x92x1x1]
%onnx::Conv_737[FLOAT, 184x1x3x3]
%onnx::Conv_740[FLOAT, 184x92x1x1]
%onnx::Conv_743[FLOAT, 1104x184x1x1]
%onnx::Conv_744[FLOAT, 1104]
%onnx::Conv_746[FLOAT, 1104x1x3x3]
%onnx::Conv_749[FLOAT, 184x1104x1x1]
%onnx::Conv_752[FLOAT, 184x184x1x1]
%onnx::Conv_755[FLOAT, 184x1x3x3]
%onnx::Conv_758[FLOAT, 184x184x1x1]
%onnx::Conv_761[FLOAT, 184x184x1x1]
%onnx::Conv_764[FLOAT, 184x1x5x5]
%onnx::Conv_767[FLOAT, 352x184x1x1]
%onnx::Conv_768[FLOAT, 352]
%onnx::Conv_770[FLOAT, 1504x352x1x1]
%onnx::Conv_771[FLOAT, 1504]
) {
%onnx::Conv_765 = Identity(%onnx::Conv_732)
%onnx::Conv_762 = Identity(%onnx::Conv_732)
%onnx::Conv_759 = Identity(%onnx::Conv_732)
%onnx::Conv_756 = Identity(%onnx::Conv_732)
%onnx::Conv_753 = Identity(%onnx::Conv_732)
%onnx::Conv_750 = Identity(%onnx::Conv_732)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_732)
%onnx::Conv_735 = Identity(%onnx::Conv_732)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_714)
%onnx::Conv_717 = Identity(%onnx::Conv_714)
%onnx::Conv_711 = Identity(%onnx::Conv_678)
%onnx::Conv_708 = Identity(%onnx::Conv_678)
%onnx::Conv_705 = Identity(%onnx::Conv_678)
%onnx::Conv_702 = Identity(%onnx::Conv_663)
%onnx::Conv_699 = Identity(%onnx::Conv_663)
%onnx::Conv_696 = Identity(%onnx::Conv_678)
%onnx::Conv_693 = Identity(%onnx::Conv_678)
%onnx::Conv_690 = Identity(%onnx::Conv_678)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_675 = Identity(%onnx::Conv_663)
%onnx::Conv_672 = Identity(%onnx::Conv_663)
%onnx::Conv_669 = Identity(%onnx::Conv_660)
%onnx::Conv_666 = Identity(%onnx::Conv_663)
%onnx::Conv_657 = Identity(%onnx::Conv_624)
%onnx::Conv_654 = Identity(%onnx::Conv_624)
%onnx::Conv_651 = Identity(%onnx::Conv_624)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%onnx::Conv_642 = Identity(%onnx::Conv_624)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%onnx::Conv_633 = Identity(%onnx::Conv_624)
%onnx::Conv_630 = Identity(%onnx::Conv_624)
%onnx::Conv_627 = Identity(%onnx::Conv_624)
%onnx::Conv_621 = Identity(%onnx::Conv_606)
%onnx::Conv_618 = Identity(%onnx::Conv_606)
%onnx::Conv_615 = Identity(%onnx::Conv_606)
%onnx::Conv_612 = Identity(%onnx::Conv_606)
%onnx::Conv_609 = Identity(%onnx::Conv_606)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_605, %onnx::Conv_606)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%603 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %603
}
|
val_accuracy
| 0 | 62,942,336 | 1,702,852 |
{'zcp_synflow': 68.89123942875973, 'zcp_zen': 60.675655364990234, 'zcp_epe_nas': 17.78819626127085, 'zcp_fisher': 0.06243060156702995, 'zcp_flops': 62942336.0, 'zcp_grad_norm': 19.846906661987305, 'zcp_grasp': -0.03765106201171875, 'zcp_jacov': -16.067104099565704, 'zcp_l2_norm': 543.3915405273438, 'zcp_nwot': 211.87951627228148, 'zcp_params': 1702852.0, 'zcp_plain': 0.004232973325997591, 'zcp_snip': 31.101463317871094, 'lat_1080ti_1': 0.35458223895679924, 'lat_1080ti_32': 0.45602721920098493, 'lat_1080ti_64': 0.4378435372288058, 'lat_2080ti_1': 0.40713398565806935, 'lat_2080ti_32': 0.4303627578391888, 'lat_2080ti_64': 0.4123139592532101, 'lat_essential_ph_1': 0.22641509433962265, 'lat_eyeriss': 0.42073606806887565, 'lat_fpga': 0.3213073551093914, 'lat_gold_6226': 0.2813847942575787, 'lat_gold_6240': 0.4096540455855252, 'lat_pixel2': 0.32608695652173914, 'lat_pixel3': 0.4508905900222019, 'lat_raspi4': 0.43901761968121006, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.13385826771653545, 'lat_silver_4114': 0.4337656438776283, 'lat_silver_4210r': 0.4301022399516007, 'lat_titan_rtx_1': 0.3917079702883237, 'lat_titan_rtx_32': 0.4021625290839881, 'lat_titan_rtx_64': 0.4227987303861793, 'lat_titanx_1': 0.21400265261924364, 'lat_titanx_32': 0.4202740267236182, 'lat_titanx_64': 0.4912989055643024, 'lat_titanxp_1': 0.3876397698268188, 'lat_titanxp_32': 0.43509350137408204, 'lat_titanxp_64': 0.429710255295848}
| |
FBNet_215
|
FBNet
|
215
|
215
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_657[FLOAT, 16x3x3x3]
%onnx::Conv_658[FLOAT, 16]
%onnx::Conv_660[FLOAT, 48x16x1x1]
%onnx::Conv_661[FLOAT, 48]
%onnx::Conv_663[FLOAT, 48x1x5x5]
%onnx::Conv_666[FLOAT, 16x48x1x1]
%onnx::Conv_669[FLOAT, 96x16x1x1]
%onnx::Conv_670[FLOAT, 96]
%onnx::Conv_672[FLOAT, 96x1x5x5]
%onnx::Conv_675[FLOAT, 24x96x1x1]
%onnx::Conv_676[FLOAT, 24]
%onnx::Conv_678[FLOAT, 72x24x1x1]
%onnx::Conv_679[FLOAT, 72]
%onnx::Conv_681[FLOAT, 72x1x3x3]
%onnx::Conv_684[FLOAT, 24x72x1x1]
%onnx::Conv_687[FLOAT, 24x12x1x1]
%onnx::Conv_690[FLOAT, 24x1x3x3]
%onnx::Conv_693[FLOAT, 24x12x1x1]
%onnx::Conv_696[FLOAT, 144x24x1x1]
%onnx::Conv_697[FLOAT, 144]
%onnx::Conv_699[FLOAT, 144x1x5x5]
%onnx::Conv_702[FLOAT, 24x144x1x1]
%onnx::Conv_705[FLOAT, 144x24x1x1]
%onnx::Conv_708[FLOAT, 144x1x5x5]
%onnx::Conv_711[FLOAT, 32x144x1x1]
%onnx::Conv_712[FLOAT, 32]
%onnx::Conv_714[FLOAT, 96x32x1x1]
%onnx::Conv_717[FLOAT, 96x1x5x5]
%onnx::Conv_720[FLOAT, 32x96x1x1]
%onnx::Conv_723[FLOAT, 96x32x1x1]
%onnx::Conv_726[FLOAT, 96x1x3x3]
%onnx::Conv_729[FLOAT, 32x96x1x1]
%onnx::Conv_732[FLOAT, 96x32x1x1]
%onnx::Conv_735[FLOAT, 96x1x5x5]
%onnx::Conv_738[FLOAT, 32x96x1x1]
%onnx::Conv_741[FLOAT, 32x16x1x1]
%onnx::Conv_744[FLOAT, 32x1x5x5]
%onnx::Conv_747[FLOAT, 64x16x1x1]
%onnx::Conv_748[FLOAT, 64]
%onnx::Conv_750[FLOAT, 64x64x1x1]
%onnx::Conv_753[FLOAT, 64x1x5x5]
%onnx::Conv_756[FLOAT, 64x64x1x1]
%onnx::Conv_759[FLOAT, 384x64x1x1]
%onnx::Conv_760[FLOAT, 384]
%onnx::Conv_762[FLOAT, 384x1x3x3]
%onnx::Conv_765[FLOAT, 64x384x1x1]
%onnx::Conv_768[FLOAT, 384x64x1x1]
%onnx::Conv_771[FLOAT, 384x1x3x3]
%onnx::Conv_774[FLOAT, 64x384x1x1]
%onnx::Conv_777[FLOAT, 192x64x1x1]
%onnx::Conv_778[FLOAT, 192]
%onnx::Conv_780[FLOAT, 192x1x5x5]
%onnx::Conv_783[FLOAT, 112x192x1x1]
%onnx::Conv_784[FLOAT, 112]
%onnx::Conv_786[FLOAT, 336x112x1x1]
%onnx::Conv_787[FLOAT, 336]
%onnx::Conv_789[FLOAT, 336x1x3x3]
%onnx::Conv_792[FLOAT, 112x336x1x1]
%onnx::Conv_795[FLOAT, 112x112x1x1]
%onnx::Conv_798[FLOAT, 112x1x3x3]
%onnx::Conv_801[FLOAT, 112x112x1x1]
%onnx::Conv_804[FLOAT, 672x112x1x1]
%onnx::Conv_805[FLOAT, 672]
%onnx::Conv_807[FLOAT, 672x1x5x5]
%onnx::Conv_810[FLOAT, 112x672x1x1]
%onnx::Conv_813[FLOAT, 336x112x1x1]
%onnx::Conv_816[FLOAT, 336x1x5x5]
%onnx::Conv_819[FLOAT, 184x336x1x1]
%onnx::Conv_820[FLOAT, 184]
%onnx::Conv_822[FLOAT, 552x184x1x1]
%onnx::Conv_823[FLOAT, 552]
%onnx::Conv_825[FLOAT, 552x1x5x5]
%onnx::Conv_828[FLOAT, 184x552x1x1]
%onnx::Conv_831[FLOAT, 1104x184x1x1]
%onnx::Conv_832[FLOAT, 1104]
%onnx::Conv_834[FLOAT, 1104x1x5x5]
%onnx::Conv_837[FLOAT, 184x1104x1x1]
%onnx::Conv_840[FLOAT, 1104x184x1x1]
%onnx::Conv_843[FLOAT, 1104x1x3x3]
%onnx::Conv_846[FLOAT, 184x1104x1x1]
%onnx::Conv_849[FLOAT, 552x184x1x1]
%onnx::Conv_852[FLOAT, 552x1x5x5]
%onnx::Conv_855[FLOAT, 352x552x1x1]
%onnx::Conv_856[FLOAT, 352]
%onnx::Conv_858[FLOAT, 1504x352x1x1]
%onnx::Conv_859[FLOAT, 1504]
) {
%onnx::Conv_853 = Identity(%onnx::Conv_823)
%onnx::Conv_850 = Identity(%onnx::Conv_823)
%onnx::Conv_847 = Identity(%onnx::Conv_820)
%onnx::Conv_844 = Identity(%onnx::Conv_832)
%onnx::Conv_841 = Identity(%onnx::Conv_832)
%onnx::Conv_838 = Identity(%onnx::Conv_820)
%onnx::Conv_835 = Identity(%onnx::Conv_832)
%onnx::Conv_829 = Identity(%onnx::Conv_820)
%onnx::Conv_826 = Identity(%onnx::Conv_823)
%onnx::Conv_817 = Identity(%onnx::Conv_787)
%onnx::Conv_814 = Identity(%onnx::Conv_787)
%onnx::Conv_811 = Identity(%onnx::Conv_784)
%onnx::Conv_808 = Identity(%onnx::Conv_805)
%onnx::Conv_802 = Identity(%onnx::Conv_784)
%onnx::Conv_799 = Identity(%onnx::Conv_784)
%onnx::Conv_796 = Identity(%onnx::Conv_784)
%onnx::Conv_793 = Identity(%onnx::Conv_784)
%onnx::Conv_790 = Identity(%onnx::Conv_787)
%onnx::Conv_781 = Identity(%onnx::Conv_778)
%onnx::Conv_775 = Identity(%onnx::Conv_748)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_748)
%onnx::Conv_763 = Identity(%onnx::Conv_760)
%onnx::Conv_757 = Identity(%onnx::Conv_748)
%onnx::Conv_754 = Identity(%onnx::Conv_748)
%onnx::Conv_751 = Identity(%onnx::Conv_748)
%onnx::Conv_745 = Identity(%onnx::Conv_712)
%onnx::Conv_742 = Identity(%onnx::Conv_712)
%onnx::Conv_739 = Identity(%onnx::Conv_712)
%onnx::Conv_736 = Identity(%onnx::Conv_670)
%onnx::Conv_733 = Identity(%onnx::Conv_670)
%onnx::Conv_730 = Identity(%onnx::Conv_712)
%onnx::Conv_727 = Identity(%onnx::Conv_670)
%onnx::Conv_724 = Identity(%onnx::Conv_670)
%onnx::Conv_721 = Identity(%onnx::Conv_712)
%onnx::Conv_718 = Identity(%onnx::Conv_670)
%onnx::Conv_715 = Identity(%onnx::Conv_670)
%onnx::Conv_709 = Identity(%onnx::Conv_697)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_676)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%onnx::Conv_694 = Identity(%onnx::Conv_676)
%onnx::Conv_691 = Identity(%onnx::Conv_676)
%onnx::Conv_688 = Identity(%onnx::Conv_676)
%onnx::Conv_685 = Identity(%onnx::Conv_676)
%onnx::Conv_682 = Identity(%onnx::Conv_679)
%onnx::Conv_673 = Identity(%onnx::Conv_670)
%onnx::Conv_667 = Identity(%onnx::Conv_658)
%onnx::Conv_664 = Identity(%onnx::Conv_661)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_657, %onnx::Conv_658)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%655 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %655
}
|
val_accuracy
| 0 | 103,206,272 | 2,686,340 |
{'zcp_synflow': 87.74283618457281, 'zcp_zen': 80.14239501953125, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.22733767330646515, 'zcp_flops': 103206272.0, 'zcp_grad_norm': 33.849708557128906, 'zcp_grasp': -0.22574234008789062, 'zcp_jacov': -16.070067988451296, 'zcp_l2_norm': 780.2084350585938, 'zcp_nwot': 218.82119817154987, 'zcp_params': 2686340.0, 'zcp_plain': 0.00044517021160572767, 'zcp_snip': 60.87909698486328, 'lat_1080ti_1': 0.7278090144210537, 'lat_1080ti_32': 0.7484350182422972, 'lat_1080ti_64': 0.7856610407546192, 'lat_2080ti_1': 0.7362431793668043, 'lat_2080ti_32': 0.7483516721226852, 'lat_2080ti_64': 0.7432722737995704, 'lat_essential_ph_1': 0.49056603773584906, 'lat_eyeriss': 0.883290851411072, 'lat_fpga': 0.8785560909963487, 'lat_gold_6226': 0.6789786190922874, 'lat_gold_6240': 0.8331598291548048, 'lat_pixel2': 0.5652173913043478, 'lat_pixel3': 0.8751792591946917, 'lat_raspi4': 0.9411860112139323, 'lat_samsung_a50': 0.37894736842105264, 'lat_samsung_s7': 0.31496062992125984, 'lat_silver_4114': 0.9802388233447964, 'lat_silver_4210r': 0.8690577196521457, 'lat_titan_rtx_1': 0.7101119037510447, 'lat_titan_rtx_32': 0.7222060969732289, 'lat_titan_rtx_64': 0.7788220309013405, 'lat_titanx_1': 0.3968494561959999, 'lat_titanx_32': 0.7891383367981303, 'lat_titanx_64': 0.7703724921334476, 'lat_titanxp_1': 0.6750204223717885, 'lat_titanxp_32': 0.7901579021116514, 'lat_titanxp_64': 0.7812720876127708}
| |
FBNet_4881
|
FBNet
|
4881
|
4881
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_614[FLOAT, 16x3x3x3]
%onnx::Conv_615[FLOAT, 16]
%onnx::Conv_617[FLOAT, 16x8x1x1]
%onnx::Conv_620[FLOAT, 16x1x5x5]
%onnx::Conv_623[FLOAT, 16x8x1x1]
%onnx::Conv_626[FLOAT, 96x16x1x1]
%onnx::Conv_627[FLOAT, 96]
%onnx::Conv_629[FLOAT, 96x1x5x5]
%onnx::Conv_632[FLOAT, 24x96x1x1]
%onnx::Conv_633[FLOAT, 24]
%onnx::Conv_635[FLOAT, 144x24x1x1]
%onnx::Conv_636[FLOAT, 144]
%onnx::Conv_638[FLOAT, 144x1x5x5]
%onnx::Conv_641[FLOAT, 24x144x1x1]
%onnx::Conv_644[FLOAT, 24x24x1x1]
%onnx::Conv_647[FLOAT, 24x1x3x3]
%onnx::Conv_650[FLOAT, 24x24x1x1]
%onnx::Conv_653[FLOAT, 24x24x1x1]
%onnx::Conv_656[FLOAT, 24x1x3x3]
%onnx::Conv_659[FLOAT, 24x24x1x1]
%onnx::Conv_662[FLOAT, 72x24x1x1]
%onnx::Conv_663[FLOAT, 72]
%onnx::Conv_665[FLOAT, 72x1x5x5]
%onnx::Conv_668[FLOAT, 32x72x1x1]
%onnx::Conv_669[FLOAT, 32]
%onnx::Conv_671[FLOAT, 32x16x1x1]
%onnx::Conv_674[FLOAT, 32x1x5x5]
%onnx::Conv_677[FLOAT, 32x16x1x1]
%onnx::Conv_680[FLOAT, 96x32x1x1]
%onnx::Conv_683[FLOAT, 96x1x3x3]
%onnx::Conv_686[FLOAT, 32x96x1x1]
%onnx::Conv_689[FLOAT, 96x32x1x1]
%onnx::Conv_692[FLOAT, 96x1x5x5]
%onnx::Conv_695[FLOAT, 64x96x1x1]
%onnx::Conv_696[FLOAT, 64]
%onnx::Conv_698[FLOAT, 192x64x1x1]
%onnx::Conv_699[FLOAT, 192]
%onnx::Conv_701[FLOAT, 192x1x3x3]
%onnx::Conv_704[FLOAT, 64x192x1x1]
%onnx::Conv_707[FLOAT, 192x64x1x1]
%onnx::Conv_710[FLOAT, 192x1x3x3]
%onnx::Conv_713[FLOAT, 64x192x1x1]
%onnx::Conv_716[FLOAT, 192x64x1x1]
%onnx::Conv_719[FLOAT, 192x1x3x3]
%onnx::Conv_722[FLOAT, 112x192x1x1]
%onnx::Conv_723[FLOAT, 112]
%onnx::Conv_725[FLOAT, 336x112x1x1]
%onnx::Conv_726[FLOAT, 336]
%onnx::Conv_728[FLOAT, 336x1x3x3]
%onnx::Conv_731[FLOAT, 112x336x1x1]
%onnx::Conv_734[FLOAT, 112x56x1x1]
%onnx::Conv_737[FLOAT, 112x1x3x3]
%onnx::Conv_740[FLOAT, 112x56x1x1]
%onnx::Conv_743[FLOAT, 336x112x1x1]
%onnx::Conv_746[FLOAT, 336x1x3x3]
%onnx::Conv_749[FLOAT, 112x336x1x1]
%onnx::Conv_752[FLOAT, 672x112x1x1]
%onnx::Conv_753[FLOAT, 672]
%onnx::Conv_755[FLOAT, 672x1x5x5]
%onnx::Conv_758[FLOAT, 184x672x1x1]
%onnx::Conv_759[FLOAT, 184]
%onnx::Conv_761[FLOAT, 184x92x1x1]
%onnx::Conv_764[FLOAT, 184x1x5x5]
%onnx::Conv_767[FLOAT, 184x92x1x1]
%onnx::Conv_770[FLOAT, 552x184x1x1]
%onnx::Conv_771[FLOAT, 552]
%onnx::Conv_773[FLOAT, 552x1x3x3]
%onnx::Conv_776[FLOAT, 184x552x1x1]
%onnx::Conv_779[FLOAT, 552x184x1x1]
%onnx::Conv_782[FLOAT, 552x1x3x3]
%onnx::Conv_785[FLOAT, 352x552x1x1]
%onnx::Conv_786[FLOAT, 352]
%onnx::Conv_788[FLOAT, 1504x352x1x1]
%onnx::Conv_789[FLOAT, 1504]
) {
%onnx::Conv_783 = Identity(%onnx::Conv_771)
%onnx::Conv_780 = Identity(%onnx::Conv_771)
%onnx::Conv_777 = Identity(%onnx::Conv_759)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_768 = Identity(%onnx::Conv_759)
%onnx::Conv_765 = Identity(%onnx::Conv_759)
%onnx::Conv_762 = Identity(%onnx::Conv_759)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_750 = Identity(%onnx::Conv_723)
%onnx::Conv_747 = Identity(%onnx::Conv_726)
%onnx::Conv_744 = Identity(%onnx::Conv_726)
%onnx::Conv_741 = Identity(%onnx::Conv_723)
%onnx::Conv_738 = Identity(%onnx::Conv_723)
%onnx::Conv_735 = Identity(%onnx::Conv_723)
%onnx::Conv_732 = Identity(%onnx::Conv_723)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_720 = Identity(%onnx::Conv_699)
%onnx::Conv_717 = Identity(%onnx::Conv_699)
%onnx::Conv_714 = Identity(%onnx::Conv_696)
%onnx::Conv_711 = Identity(%onnx::Conv_699)
%onnx::Conv_708 = Identity(%onnx::Conv_699)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%onnx::Conv_693 = Identity(%onnx::Conv_627)
%onnx::Conv_690 = Identity(%onnx::Conv_627)
%onnx::Conv_687 = Identity(%onnx::Conv_669)
%onnx::Conv_684 = Identity(%onnx::Conv_627)
%onnx::Conv_681 = Identity(%onnx::Conv_627)
%onnx::Conv_678 = Identity(%onnx::Conv_669)
%onnx::Conv_675 = Identity(%onnx::Conv_669)
%onnx::Conv_672 = Identity(%onnx::Conv_669)
%onnx::Conv_666 = Identity(%onnx::Conv_663)
%onnx::Conv_660 = Identity(%onnx::Conv_633)
%onnx::Conv_657 = Identity(%onnx::Conv_633)
%onnx::Conv_654 = Identity(%onnx::Conv_633)
%onnx::Conv_651 = Identity(%onnx::Conv_633)
%onnx::Conv_648 = Identity(%onnx::Conv_633)
%onnx::Conv_645 = Identity(%onnx::Conv_633)
%onnx::Conv_642 = Identity(%onnx::Conv_633)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%onnx::Conv_630 = Identity(%onnx::Conv_627)
%onnx::Conv_624 = Identity(%onnx::Conv_615)
%onnx::Conv_621 = Identity(%onnx::Conv_615)
%onnx::Conv_618 = Identity(%onnx::Conv_615)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_614, %onnx::Conv_615)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%612 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %612
}
|
val_accuracy
| 0 | 71,637,888 | 1,770,356 |
{'zcp_synflow': 73.05878391459093, 'zcp_zen': 64.4052734375, 'zcp_epe_nas': 8.67919470167045, 'zcp_fisher': 0.11168713867664337, 'zcp_flops': 71637888.0, 'zcp_grad_norm': 24.438800811767578, 'zcp_grasp': -0.039669036865234375, 'zcp_jacov': -16.057331161192018, 'zcp_l2_norm': 589.5195922851562, 'zcp_nwot': 213.4385291795488, 'zcp_params': 1770356.0, 'zcp_plain': 0.004999706055969, 'zcp_snip': 44.09947967529297, 'lat_1080ti_1': 0.49906462527116124, 'lat_1080ti_32': 0.5318884849845732, 'lat_1080ti_64': 0.48279616118374724, 'lat_2080ti_1': 0.5099098753877871, 'lat_2080ti_32': 0.5690456946295295, 'lat_2080ti_64': 0.470292301537308, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.4431078429574752, 'lat_fpga': 0.4492722073244342, 'lat_gold_6226': 0.3549614922997302, 'lat_gold_6240': 0.40362350608540054, 'lat_pixel2': 0.45652173913043476, 'lat_pixel3': 0.4857779285147163, 'lat_raspi4': 0.5358245659120392, 'lat_samsung_a50': 0.18947368421052632, 'lat_samsung_s7': 0.15748031496062992, 'lat_silver_4114': 0.4677936074873993, 'lat_silver_4210r': 0.4538097141946063, 'lat_titan_rtx_1': 0.47414656364031904, 'lat_titan_rtx_32': 0.5323113223922453, 'lat_titan_rtx_64': 0.4847109451991754, 'lat_titanx_1': 0.2508223799831015, 'lat_titanx_32': 0.4972989808014636, 'lat_titanx_64': 0.5243653332706637, 'lat_titanxp_1': 0.43861788054345774, 'lat_titanxp_32': 0.5250025118291695, 'lat_titanxp_64': 0.48728319619886784}
| |
FBNet_3103
|
FBNet
|
3103
|
3103
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_714[FLOAT, 16x3x3x3]
%onnx::Conv_715[FLOAT, 16]
%onnx::Conv_717[FLOAT, 48x16x1x1]
%onnx::Conv_718[FLOAT, 48]
%onnx::Conv_720[FLOAT, 48x1x3x3]
%onnx::Conv_723[FLOAT, 16x48x1x1]
%onnx::Conv_726[FLOAT, 16x16x1x1]
%onnx::Conv_729[FLOAT, 16x1x3x3]
%onnx::Conv_732[FLOAT, 24x16x1x1]
%onnx::Conv_733[FLOAT, 24]
%onnx::Conv_735[FLOAT, 24x12x1x1]
%onnx::Conv_738[FLOAT, 24x1x3x3]
%onnx::Conv_741[FLOAT, 24x12x1x1]
%onnx::Conv_744[FLOAT, 72x24x1x1]
%onnx::Conv_745[FLOAT, 72]
%onnx::Conv_747[FLOAT, 72x1x3x3]
%onnx::Conv_750[FLOAT, 24x72x1x1]
%onnx::Conv_753[FLOAT, 24x24x1x1]
%onnx::Conv_756[FLOAT, 24x1x5x5]
%onnx::Conv_759[FLOAT, 24x24x1x1]
%onnx::Conv_762[FLOAT, 72x24x1x1]
%onnx::Conv_765[FLOAT, 72x1x5x5]
%onnx::Conv_768[FLOAT, 32x72x1x1]
%onnx::Conv_769[FLOAT, 32]
%onnx::Conv_771[FLOAT, 96x32x1x1]
%onnx::Conv_772[FLOAT, 96]
%onnx::Conv_774[FLOAT, 96x1x5x5]
%onnx::Conv_777[FLOAT, 32x96x1x1]
%onnx::Conv_780[FLOAT, 192x32x1x1]
%onnx::Conv_781[FLOAT, 192]
%onnx::Conv_783[FLOAT, 192x1x3x3]
%onnx::Conv_786[FLOAT, 32x192x1x1]
%onnx::Conv_789[FLOAT, 32x16x1x1]
%onnx::Conv_792[FLOAT, 32x1x5x5]
%onnx::Conv_795[FLOAT, 32x16x1x1]
%onnx::Conv_798[FLOAT, 96x32x1x1]
%onnx::Conv_801[FLOAT, 96x1x3x3]
%onnx::Conv_804[FLOAT, 64x96x1x1]
%onnx::Conv_805[FLOAT, 64]
%onnx::Conv_807[FLOAT, 192x64x1x1]
%onnx::Conv_810[FLOAT, 192x1x3x3]
%onnx::Conv_813[FLOAT, 64x192x1x1]
%onnx::Conv_816[FLOAT, 64x64x1x1]
%onnx::Conv_819[FLOAT, 64x1x5x5]
%onnx::Conv_822[FLOAT, 64x64x1x1]
%onnx::Conv_825[FLOAT, 64x64x1x1]
%onnx::Conv_828[FLOAT, 64x1x5x5]
%onnx::Conv_831[FLOAT, 64x64x1x1]
%onnx::Conv_834[FLOAT, 64x64x1x1]
%onnx::Conv_837[FLOAT, 64x1x3x3]
%onnx::Conv_840[FLOAT, 112x64x1x1]
%onnx::Conv_841[FLOAT, 112]
%onnx::Conv_843[FLOAT, 112x112x1x1]
%onnx::Conv_846[FLOAT, 112x1x3x3]
%onnx::Conv_849[FLOAT, 112x112x1x1]
%onnx::Conv_852[FLOAT, 112x56x1x1]
%onnx::Conv_855[FLOAT, 112x1x3x3]
%onnx::Conv_858[FLOAT, 112x56x1x1]
%onnx::Conv_861[FLOAT, 112x112x1x1]
%onnx::Conv_864[FLOAT, 112x1x5x5]
%onnx::Conv_867[FLOAT, 112x112x1x1]
%onnx::Conv_870[FLOAT, 184x112x1x1]
%onnx::Conv_871[FLOAT, 184]
%onnx::Conv_873[FLOAT, 184x92x1x1]
%onnx::Conv_876[FLOAT, 184x1x3x3]
%onnx::Conv_879[FLOAT, 184x92x1x1]
%onnx::Conv_882[FLOAT, 552x184x1x1]
%onnx::Conv_883[FLOAT, 552]
%onnx::Conv_885[FLOAT, 552x1x5x5]
%onnx::Conv_888[FLOAT, 184x552x1x1]
%onnx::Conv_891[FLOAT, 184x92x1x1]
%onnx::Conv_894[FLOAT, 184x1x5x5]
%onnx::Conv_897[FLOAT, 184x92x1x1]
%onnx::Conv_900[FLOAT, 184x92x1x1]
%onnx::Conv_903[FLOAT, 184x1x3x3]
%onnx::Conv_906[FLOAT, 352x92x1x1]
%onnx::Conv_907[FLOAT, 352]
%onnx::Conv_909[FLOAT, 1504x352x1x1]
%onnx::Conv_910[FLOAT, 1504]
) {
%onnx::Conv_904 = Identity(%onnx::Conv_871)
%onnx::Conv_901 = Identity(%onnx::Conv_871)
%onnx::Conv_898 = Identity(%onnx::Conv_871)
%onnx::Conv_895 = Identity(%onnx::Conv_871)
%onnx::Conv_892 = Identity(%onnx::Conv_871)
%onnx::Conv_889 = Identity(%onnx::Conv_871)
%onnx::Conv_886 = Identity(%onnx::Conv_883)
%onnx::Conv_880 = Identity(%onnx::Conv_871)
%onnx::Conv_877 = Identity(%onnx::Conv_871)
%onnx::Conv_874 = Identity(%onnx::Conv_871)
%onnx::Conv_868 = Identity(%onnx::Conv_841)
%onnx::Conv_865 = Identity(%onnx::Conv_841)
%onnx::Conv_862 = Identity(%onnx::Conv_841)
%onnx::Conv_859 = Identity(%onnx::Conv_841)
%onnx::Conv_856 = Identity(%onnx::Conv_841)
%onnx::Conv_853 = Identity(%onnx::Conv_841)
%onnx::Conv_850 = Identity(%onnx::Conv_841)
%onnx::Conv_847 = Identity(%onnx::Conv_841)
%onnx::Conv_844 = Identity(%onnx::Conv_841)
%onnx::Conv_838 = Identity(%onnx::Conv_805)
%onnx::Conv_835 = Identity(%onnx::Conv_805)
%onnx::Conv_832 = Identity(%onnx::Conv_805)
%onnx::Conv_829 = Identity(%onnx::Conv_805)
%onnx::Conv_826 = Identity(%onnx::Conv_805)
%onnx::Conv_823 = Identity(%onnx::Conv_805)
%onnx::Conv_820 = Identity(%onnx::Conv_805)
%onnx::Conv_817 = Identity(%onnx::Conv_805)
%onnx::Conv_814 = Identity(%onnx::Conv_805)
%onnx::Conv_811 = Identity(%onnx::Conv_781)
%onnx::Conv_808 = Identity(%onnx::Conv_781)
%onnx::Conv_802 = Identity(%onnx::Conv_772)
%onnx::Conv_799 = Identity(%onnx::Conv_772)
%onnx::Conv_796 = Identity(%onnx::Conv_769)
%onnx::Conv_793 = Identity(%onnx::Conv_769)
%onnx::Conv_790 = Identity(%onnx::Conv_769)
%onnx::Conv_787 = Identity(%onnx::Conv_769)
%onnx::Conv_784 = Identity(%onnx::Conv_781)
%onnx::Conv_778 = Identity(%onnx::Conv_769)
%onnx::Conv_775 = Identity(%onnx::Conv_772)
%onnx::Conv_766 = Identity(%onnx::Conv_745)
%onnx::Conv_763 = Identity(%onnx::Conv_745)
%onnx::Conv_760 = Identity(%onnx::Conv_733)
%onnx::Conv_757 = Identity(%onnx::Conv_733)
%onnx::Conv_754 = Identity(%onnx::Conv_733)
%onnx::Conv_751 = Identity(%onnx::Conv_733)
%onnx::Conv_748 = Identity(%onnx::Conv_745)
%onnx::Conv_742 = Identity(%onnx::Conv_733)
%onnx::Conv_739 = Identity(%onnx::Conv_733)
%onnx::Conv_736 = Identity(%onnx::Conv_733)
%onnx::Conv_730 = Identity(%onnx::Conv_715)
%onnx::Conv_727 = Identity(%onnx::Conv_715)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_714, %onnx::Conv_715)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_882, %onnx::Conv_883)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_885, %onnx::Conv_886)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_888, %onnx::Conv_889)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_891, %onnx::Conv_892)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_894, %onnx::Conv_895)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_897, %onnx::Conv_898)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_900, %onnx::Conv_901)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_903, %onnx::Conv_904)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_906, %onnx::Conv_907)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_909, %onnx::Conv_910)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%712 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %712
}
|
val_accuracy
| 0 | 45,311,744 | 1,234,868 |
{'zcp_synflow': 79.45354202616066, 'zcp_zen': 66.99819946289062, 'zcp_epe_nas': 8.237589104597992, 'zcp_fisher': 0.12265849858522415, 'zcp_flops': 45311744.0, 'zcp_grad_norm': 26.906280517578125, 'zcp_grasp': 0.39290618896484375, 'zcp_jacov': -16.05804585714609, 'zcp_l2_norm': 556.679443359375, 'zcp_nwot': 208.90060527715227, 'zcp_params': 1234868.0, 'zcp_plain': -0.006799100898206234, 'zcp_snip': 42.719993591308594, 'lat_1080ti_1': 0.7612620459363536, 'lat_1080ti_32': 0.5752190921651744, 'lat_1080ti_64': 0.40591874859974303, 'lat_2080ti_1': 0.7880206356559969, 'lat_2080ti_32': 0.6196166253072888, 'lat_2080ti_64': 0.42348284345753984, 'lat_essential_ph_1': 0.20754716981132076, 'lat_eyeriss': 0.19704582546348076, 'lat_fpga': 0.16787223305197937, 'lat_gold_6226': 0.11405782484969057, 'lat_gold_6240': 0.4635892862087895, 'lat_pixel2': 0.13043478260869565, 'lat_pixel3': 0.1814656100078242, 'lat_raspi4': 0.19982157848705126, 'lat_samsung_a50': 0.09473684210526316, 'lat_samsung_s7': 0.08661417322834646, 'lat_silver_4114': 0.39071098124802733, 'lat_silver_4210r': 0.41089390854392677, 'lat_titan_rtx_1': 0.7424944641376541, 'lat_titan_rtx_32': 0.6155652136023386, 'lat_titan_rtx_64': 0.45722108923947374, 'lat_titanx_1': 0.3918848576331056, 'lat_titanx_32': 0.4926237837609285, 'lat_titanx_64': 0.36713296254783606, 'lat_titanxp_1': 0.6816964711658514, 'lat_titanxp_32': 0.5592127094418384, 'lat_titanxp_64': 0.40495637292818837}
| |
FBNet_2065
|
FBNet
|
2065
|
2065
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_714[FLOAT, 16x3x3x3]
%onnx::Conv_715[FLOAT, 16]
%onnx::Conv_717[FLOAT, 48x16x1x1]
%onnx::Conv_718[FLOAT, 48]
%onnx::Conv_720[FLOAT, 48x1x3x3]
%onnx::Conv_723[FLOAT, 16x48x1x1]
%onnx::Conv_726[FLOAT, 16x16x1x1]
%onnx::Conv_729[FLOAT, 16x1x3x3]
%onnx::Conv_732[FLOAT, 24x16x1x1]
%onnx::Conv_733[FLOAT, 24]
%onnx::Conv_735[FLOAT, 144x24x1x1]
%onnx::Conv_736[FLOAT, 144]
%onnx::Conv_738[FLOAT, 144x1x3x3]
%onnx::Conv_741[FLOAT, 24x144x1x1]
%onnx::Conv_744[FLOAT, 24x12x1x1]
%onnx::Conv_747[FLOAT, 24x1x3x3]
%onnx::Conv_750[FLOAT, 24x12x1x1]
%onnx::Conv_753[FLOAT, 144x24x1x1]
%onnx::Conv_756[FLOAT, 144x1x3x3]
%onnx::Conv_759[FLOAT, 24x144x1x1]
%onnx::Conv_762[FLOAT, 24x12x1x1]
%onnx::Conv_765[FLOAT, 24x1x5x5]
%onnx::Conv_768[FLOAT, 32x12x1x1]
%onnx::Conv_769[FLOAT, 32]
%onnx::Conv_771[FLOAT, 96x32x1x1]
%onnx::Conv_772[FLOAT, 96]
%onnx::Conv_774[FLOAT, 96x1x5x5]
%onnx::Conv_777[FLOAT, 32x96x1x1]
%onnx::Conv_780[FLOAT, 32x16x1x1]
%onnx::Conv_783[FLOAT, 32x1x3x3]
%onnx::Conv_786[FLOAT, 32x16x1x1]
%onnx::Conv_789[FLOAT, 96x32x1x1]
%onnx::Conv_792[FLOAT, 96x1x3x3]
%onnx::Conv_795[FLOAT, 32x96x1x1]
%onnx::Conv_798[FLOAT, 32x16x1x1]
%onnx::Conv_801[FLOAT, 32x1x3x3]
%onnx::Conv_804[FLOAT, 64x16x1x1]
%onnx::Conv_805[FLOAT, 64]
%onnx::Conv_807[FLOAT, 64x32x1x1]
%onnx::Conv_810[FLOAT, 64x1x3x3]
%onnx::Conv_813[FLOAT, 64x32x1x1]
%onnx::Conv_816[FLOAT, 64x64x1x1]
%onnx::Conv_819[FLOAT, 64x1x5x5]
%onnx::Conv_822[FLOAT, 64x64x1x1]
%onnx::Conv_825[FLOAT, 64x64x1x1]
%onnx::Conv_828[FLOAT, 64x1x3x3]
%onnx::Conv_831[FLOAT, 112x64x1x1]
%onnx::Conv_832[FLOAT, 112]
%onnx::Conv_834[FLOAT, 336x112x1x1]
%onnx::Conv_835[FLOAT, 336]
%onnx::Conv_837[FLOAT, 336x1x3x3]
%onnx::Conv_840[FLOAT, 112x336x1x1]
%onnx::Conv_843[FLOAT, 112x112x1x1]
%onnx::Conv_846[FLOAT, 112x1x5x5]
%onnx::Conv_849[FLOAT, 112x112x1x1]
%onnx::Conv_852[FLOAT, 112x56x1x1]
%onnx::Conv_855[FLOAT, 112x1x5x5]
%onnx::Conv_858[FLOAT, 112x56x1x1]
%onnx::Conv_861[FLOAT, 112x56x1x1]
%onnx::Conv_864[FLOAT, 112x1x5x5]
%onnx::Conv_867[FLOAT, 184x56x1x1]
%onnx::Conv_868[FLOAT, 184]
%onnx::Conv_870[FLOAT, 184x92x1x1]
%onnx::Conv_873[FLOAT, 184x1x3x3]
%onnx::Conv_876[FLOAT, 184x92x1x1]
%onnx::Conv_879[FLOAT, 1104x184x1x1]
%onnx::Conv_880[FLOAT, 1104]
%onnx::Conv_882[FLOAT, 1104x1x5x5]
%onnx::Conv_885[FLOAT, 184x1104x1x1]
%onnx::Conv_888[FLOAT, 184x184x1x1]
%onnx::Conv_891[FLOAT, 184x1x5x5]
%onnx::Conv_894[FLOAT, 352x184x1x1]
%onnx::Conv_895[FLOAT, 352]
%onnx::Conv_897[FLOAT, 1504x352x1x1]
%onnx::Conv_898[FLOAT, 1504]
) {
%onnx::Conv_892 = Identity(%onnx::Conv_868)
%onnx::Conv_889 = Identity(%onnx::Conv_868)
%onnx::Conv_886 = Identity(%onnx::Conv_868)
%onnx::Conv_883 = Identity(%onnx::Conv_880)
%onnx::Conv_877 = Identity(%onnx::Conv_868)
%onnx::Conv_874 = Identity(%onnx::Conv_868)
%onnx::Conv_871 = Identity(%onnx::Conv_868)
%onnx::Conv_865 = Identity(%onnx::Conv_832)
%onnx::Conv_862 = Identity(%onnx::Conv_832)
%onnx::Conv_859 = Identity(%onnx::Conv_832)
%onnx::Conv_856 = Identity(%onnx::Conv_832)
%onnx::Conv_853 = Identity(%onnx::Conv_832)
%onnx::Conv_850 = Identity(%onnx::Conv_832)
%onnx::Conv_847 = Identity(%onnx::Conv_832)
%onnx::Conv_844 = Identity(%onnx::Conv_832)
%onnx::Conv_841 = Identity(%onnx::Conv_832)
%onnx::Conv_838 = Identity(%onnx::Conv_835)
%onnx::Conv_829 = Identity(%onnx::Conv_805)
%onnx::Conv_826 = Identity(%onnx::Conv_805)
%onnx::Conv_823 = Identity(%onnx::Conv_805)
%onnx::Conv_820 = Identity(%onnx::Conv_805)
%onnx::Conv_817 = Identity(%onnx::Conv_805)
%onnx::Conv_814 = Identity(%onnx::Conv_805)
%onnx::Conv_811 = Identity(%onnx::Conv_805)
%onnx::Conv_808 = Identity(%onnx::Conv_805)
%onnx::Conv_802 = Identity(%onnx::Conv_769)
%onnx::Conv_799 = Identity(%onnx::Conv_769)
%onnx::Conv_796 = Identity(%onnx::Conv_769)
%onnx::Conv_793 = Identity(%onnx::Conv_772)
%onnx::Conv_790 = Identity(%onnx::Conv_772)
%onnx::Conv_787 = Identity(%onnx::Conv_769)
%onnx::Conv_784 = Identity(%onnx::Conv_769)
%onnx::Conv_781 = Identity(%onnx::Conv_769)
%onnx::Conv_778 = Identity(%onnx::Conv_769)
%onnx::Conv_775 = Identity(%onnx::Conv_772)
%onnx::Conv_766 = Identity(%onnx::Conv_733)
%onnx::Conv_763 = Identity(%onnx::Conv_733)
%onnx::Conv_760 = Identity(%onnx::Conv_733)
%onnx::Conv_757 = Identity(%onnx::Conv_736)
%onnx::Conv_754 = Identity(%onnx::Conv_736)
%onnx::Conv_751 = Identity(%onnx::Conv_733)
%onnx::Conv_748 = Identity(%onnx::Conv_733)
%onnx::Conv_745 = Identity(%onnx::Conv_733)
%onnx::Conv_742 = Identity(%onnx::Conv_733)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_730 = Identity(%onnx::Conv_715)
%onnx::Conv_727 = Identity(%onnx::Conv_715)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_714, %onnx::Conv_715)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_882, %onnx::Conv_883)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_885, %onnx::Conv_886)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_888, %onnx::Conv_889)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_891, %onnx::Conv_892)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_894, %onnx::Conv_895)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_897, %onnx::Conv_898)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%712 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %712
}
|
val_accuracy
| 0 | 57,046,144 | 1,478,644 |
{'zcp_synflow': 69.98533079799631, 'zcp_zen': 61.26506805419922, 'zcp_epe_nas': 6.346881732588393, 'zcp_fisher': 0.12028071284294128, 'zcp_flops': 57046144.0, 'zcp_grad_norm': 21.705698013305664, 'zcp_grasp': 0.20720672607421875, 'zcp_jacov': -16.057315180296474, 'zcp_l2_norm': 524.2155151367188, 'zcp_nwot': 213.33053739902977, 'zcp_params': 1478644.0, 'zcp_plain': 0.006678152829408646, 'zcp_snip': 40.090782165527344, 'lat_1080ti_1': 0.5975517675970677, 'lat_1080ti_32': 0.6500577579822499, 'lat_1080ti_64': 0.5232484679782542, 'lat_2080ti_1': 0.6795010423971876, 'lat_2080ti_32': 0.6798671095087587, 'lat_2080ti_64': 0.5632522800577078, 'lat_essential_ph_1': 0.22641509433962265, 'lat_eyeriss': 0.3296169291988337, 'lat_fpga': 0.3598294065842724, 'lat_gold_6226': 0.17916260576294316, 'lat_gold_6240': 0.3838377428869668, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.3248326297380271, 'lat_raspi4': 0.3800536753104946, 'lat_samsung_a50': 0.1368421052631579, 'lat_samsung_s7': 0.11811023622047244, 'lat_silver_4114': 0.4305330140030712, 'lat_silver_4210r': 0.44756268508569164, 'lat_titan_rtx_1': 0.6454535535625328, 'lat_titan_rtx_32': 0.63185396312105, 'lat_titan_rtx_64': 0.6063286536273453, 'lat_titanx_1': 0.3384668246686617, 'lat_titanx_32': 0.6010223699056799, 'lat_titanx_64': 0.4811301656600229, 'lat_titanxp_1': 0.6140548387306174, 'lat_titanxp_32': 0.6402720508463273, 'lat_titanxp_64': 0.5318309962442742}
| |
FBNet_4927
|
FBNet
|
4927
|
4927
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_698[FLOAT, 16x3x3x3]
%onnx::Conv_699[FLOAT, 16]
%onnx::Conv_701[FLOAT, 16x8x1x1]
%onnx::Conv_704[FLOAT, 16x1x5x5]
%onnx::Conv_707[FLOAT, 16x8x1x1]
%onnx::Conv_710[FLOAT, 16x8x1x1]
%onnx::Conv_713[FLOAT, 16x1x3x3]
%onnx::Conv_716[FLOAT, 24x8x1x1]
%onnx::Conv_717[FLOAT, 24]
%onnx::Conv_719[FLOAT, 24x24x1x1]
%onnx::Conv_722[FLOAT, 24x1x5x5]
%onnx::Conv_725[FLOAT, 24x24x1x1]
%onnx::Conv_728[FLOAT, 24x12x1x1]
%onnx::Conv_731[FLOAT, 24x1x5x5]
%onnx::Conv_734[FLOAT, 24x12x1x1]
%onnx::Conv_737[FLOAT, 32x24x1x1]
%onnx::Conv_738[FLOAT, 32]
%onnx::Conv_740[FLOAT, 32x16x1x1]
%onnx::Conv_743[FLOAT, 32x1x3x3]
%onnx::Conv_746[FLOAT, 32x16x1x1]
%onnx::Conv_749[FLOAT, 32x16x1x1]
%onnx::Conv_752[FLOAT, 32x1x5x5]
%onnx::Conv_755[FLOAT, 32x16x1x1]
%onnx::Conv_758[FLOAT, 96x32x1x1]
%onnx::Conv_759[FLOAT, 96]
%onnx::Conv_761[FLOAT, 96x1x5x5]
%onnx::Conv_764[FLOAT, 32x96x1x1]
%onnx::Conv_767[FLOAT, 96x32x1x1]
%onnx::Conv_770[FLOAT, 96x1x5x5]
%onnx::Conv_773[FLOAT, 64x96x1x1]
%onnx::Conv_774[FLOAT, 64]
%onnx::Conv_776[FLOAT, 192x64x1x1]
%onnx::Conv_777[FLOAT, 192]
%onnx::Conv_779[FLOAT, 192x1x3x3]
%onnx::Conv_782[FLOAT, 64x192x1x1]
%onnx::Conv_785[FLOAT, 192x64x1x1]
%onnx::Conv_788[FLOAT, 192x1x5x5]
%onnx::Conv_791[FLOAT, 64x192x1x1]
%onnx::Conv_794[FLOAT, 64x32x1x1]
%onnx::Conv_797[FLOAT, 64x1x5x5]
%onnx::Conv_800[FLOAT, 64x32x1x1]
%onnx::Conv_803[FLOAT, 112x64x1x1]
%onnx::Conv_804[FLOAT, 112]
%onnx::Conv_806[FLOAT, 112x56x1x1]
%onnx::Conv_809[FLOAT, 112x1x3x3]
%onnx::Conv_812[FLOAT, 112x56x1x1]
%onnx::Conv_815[FLOAT, 112x56x1x1]
%onnx::Conv_818[FLOAT, 112x1x3x3]
%onnx::Conv_821[FLOAT, 112x56x1x1]
%onnx::Conv_824[FLOAT, 112x56x1x1]
%onnx::Conv_827[FLOAT, 112x1x5x5]
%onnx::Conv_830[FLOAT, 112x56x1x1]
%onnx::Conv_833[FLOAT, 112x112x1x1]
%onnx::Conv_836[FLOAT, 112x1x5x5]
%onnx::Conv_839[FLOAT, 184x112x1x1]
%onnx::Conv_840[FLOAT, 184]
%onnx::Conv_842[FLOAT, 552x184x1x1]
%onnx::Conv_843[FLOAT, 552]
%onnx::Conv_845[FLOAT, 552x1x3x3]
%onnx::Conv_848[FLOAT, 184x552x1x1]
%onnx::Conv_851[FLOAT, 552x184x1x1]
%onnx::Conv_854[FLOAT, 552x1x3x3]
%onnx::Conv_857[FLOAT, 184x552x1x1]
%onnx::Conv_860[FLOAT, 552x184x1x1]
%onnx::Conv_863[FLOAT, 552x1x5x5]
%onnx::Conv_866[FLOAT, 352x552x1x1]
%onnx::Conv_867[FLOAT, 352]
%onnx::Conv_869[FLOAT, 1504x352x1x1]
%onnx::Conv_870[FLOAT, 1504]
) {
%onnx::Conv_864 = Identity(%onnx::Conv_843)
%onnx::Conv_861 = Identity(%onnx::Conv_843)
%onnx::Conv_858 = Identity(%onnx::Conv_840)
%onnx::Conv_855 = Identity(%onnx::Conv_843)
%onnx::Conv_852 = Identity(%onnx::Conv_843)
%onnx::Conv_849 = Identity(%onnx::Conv_840)
%onnx::Conv_846 = Identity(%onnx::Conv_843)
%onnx::Conv_837 = Identity(%onnx::Conv_804)
%onnx::Conv_834 = Identity(%onnx::Conv_804)
%onnx::Conv_831 = Identity(%onnx::Conv_804)
%onnx::Conv_828 = Identity(%onnx::Conv_804)
%onnx::Conv_825 = Identity(%onnx::Conv_804)
%onnx::Conv_822 = Identity(%onnx::Conv_804)
%onnx::Conv_819 = Identity(%onnx::Conv_804)
%onnx::Conv_816 = Identity(%onnx::Conv_804)
%onnx::Conv_813 = Identity(%onnx::Conv_804)
%onnx::Conv_810 = Identity(%onnx::Conv_804)
%onnx::Conv_807 = Identity(%onnx::Conv_804)
%onnx::Conv_801 = Identity(%onnx::Conv_774)
%onnx::Conv_798 = Identity(%onnx::Conv_774)
%onnx::Conv_795 = Identity(%onnx::Conv_774)
%onnx::Conv_792 = Identity(%onnx::Conv_774)
%onnx::Conv_789 = Identity(%onnx::Conv_777)
%onnx::Conv_786 = Identity(%onnx::Conv_777)
%onnx::Conv_783 = Identity(%onnx::Conv_774)
%onnx::Conv_780 = Identity(%onnx::Conv_777)
%onnx::Conv_771 = Identity(%onnx::Conv_759)
%onnx::Conv_768 = Identity(%onnx::Conv_759)
%onnx::Conv_765 = Identity(%onnx::Conv_738)
%onnx::Conv_762 = Identity(%onnx::Conv_759)
%onnx::Conv_756 = Identity(%onnx::Conv_738)
%onnx::Conv_753 = Identity(%onnx::Conv_738)
%onnx::Conv_750 = Identity(%onnx::Conv_738)
%onnx::Conv_747 = Identity(%onnx::Conv_738)
%onnx::Conv_744 = Identity(%onnx::Conv_738)
%onnx::Conv_741 = Identity(%onnx::Conv_738)
%onnx::Conv_735 = Identity(%onnx::Conv_717)
%onnx::Conv_732 = Identity(%onnx::Conv_717)
%onnx::Conv_729 = Identity(%onnx::Conv_717)
%onnx::Conv_726 = Identity(%onnx::Conv_717)
%onnx::Conv_723 = Identity(%onnx::Conv_717)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%onnx::Conv_714 = Identity(%onnx::Conv_699)
%onnx::Conv_711 = Identity(%onnx::Conv_699)
%onnx::Conv_708 = Identity(%onnx::Conv_699)
%onnx::Conv_705 = Identity(%onnx::Conv_699)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_698, %onnx::Conv_699)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.4/Add_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%696 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %696
}
|
val_accuracy
| 0 | 39,233,920 | 1,599,516 |
{'zcp_synflow': 68.31832768040292, 'zcp_zen': 58.89002227783203, 'zcp_epe_nas': 6.498296902732021, 'zcp_fisher': 0.07777952402830124, 'zcp_flops': 39233920.0, 'zcp_grad_norm': 18.319854736328125, 'zcp_grasp': 0.04257011413574219, 'zcp_jacov': -16.051610085744315, 'zcp_l2_norm': 508.1073303222656, 'zcp_nwot': 200.28207966542064, 'zcp_params': 1599516.0, 'zcp_plain': 0.001407412113621831, 'zcp_snip': 30.805438995361328, 'lat_1080ti_1': 0.6089757137323908, 'lat_1080ti_32': 0.39357734507604447, 'lat_1080ti_64': 0.19988437718223487, 'lat_2080ti_1': 0.5972359510568389, 'lat_2080ti_32': 0.45644278131411803, 'lat_2080ti_64': 0.2257793440086058, 'lat_essential_ph_1': 0.07547169811320754, 'lat_eyeriss': 0.09912346652485651, 'lat_fpga': 0.10192066020839315, 'lat_gold_6226': 0.1836265938177469, 'lat_gold_6240': 0.30925107999451223, 'lat_pixel2': 0.10869565217391304, 'lat_pixel3': 0.14272646072505643, 'lat_raspi4': 0.2391157296835532, 'lat_samsung_a50': 0.06315789473684211, 'lat_samsung_s7': 0.07874015748031496, 'lat_silver_4114': 0.34393746795457625, 'lat_silver_4210r': 0.34284664192154524, 'lat_titan_rtx_1': 0.5476249235565213, 'lat_titan_rtx_32': 0.4540251142369503, 'lat_titan_rtx_64': 0.265260062877221, 'lat_titanx_1': 0.28696169704533037, 'lat_titanx_32': 0.32016665765793767, 'lat_titanx_64': 0.19394205247395283, 'lat_titanxp_1': 0.5453462020357583, 'lat_titanxp_32': 0.39400704258016417, 'lat_titanxp_64': 0.21300098120417732}
| |
FBNet_1389
|
FBNet
|
1389
|
1389
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_741[FLOAT, 16x3x3x3]
%onnx::Conv_742[FLOAT, 16]
%onnx::Conv_744[FLOAT, 48x16x1x1]
%onnx::Conv_745[FLOAT, 48]
%onnx::Conv_747[FLOAT, 48x1x3x3]
%onnx::Conv_750[FLOAT, 16x48x1x1]
%onnx::Conv_753[FLOAT, 16x16x1x1]
%onnx::Conv_756[FLOAT, 16x1x3x3]
%onnx::Conv_759[FLOAT, 24x16x1x1]
%onnx::Conv_760[FLOAT, 24]
%onnx::Conv_762[FLOAT, 24x12x1x1]
%onnx::Conv_765[FLOAT, 24x1x5x5]
%onnx::Conv_768[FLOAT, 24x12x1x1]
%onnx::Conv_771[FLOAT, 24x12x1x1]
%onnx::Conv_774[FLOAT, 24x1x5x5]
%onnx::Conv_777[FLOAT, 24x12x1x1]
%onnx::Conv_780[FLOAT, 24x12x1x1]
%onnx::Conv_783[FLOAT, 24x1x5x5]
%onnx::Conv_786[FLOAT, 32x12x1x1]
%onnx::Conv_787[FLOAT, 32]
%onnx::Conv_789[FLOAT, 192x32x1x1]
%onnx::Conv_790[FLOAT, 192]
%onnx::Conv_792[FLOAT, 192x1x5x5]
%onnx::Conv_795[FLOAT, 32x192x1x1]
%onnx::Conv_798[FLOAT, 32x16x1x1]
%onnx::Conv_801[FLOAT, 32x1x3x3]
%onnx::Conv_804[FLOAT, 32x16x1x1]
%onnx::Conv_807[FLOAT, 32x32x1x1]
%onnx::Conv_810[FLOAT, 32x1x5x5]
%onnx::Conv_813[FLOAT, 32x32x1x1]
%onnx::Conv_816[FLOAT, 32x32x1x1]
%onnx::Conv_819[FLOAT, 32x1x3x3]
%onnx::Conv_822[FLOAT, 64x32x1x1]
%onnx::Conv_823[FLOAT, 64]
%onnx::Conv_825[FLOAT, 192x64x1x1]
%onnx::Conv_828[FLOAT, 192x1x3x3]
%onnx::Conv_831[FLOAT, 64x192x1x1]
%onnx::Conv_834[FLOAT, 64x64x1x1]
%onnx::Conv_837[FLOAT, 64x1x3x3]
%onnx::Conv_840[FLOAT, 64x64x1x1]
%onnx::Conv_843[FLOAT, 64x32x1x1]
%onnx::Conv_846[FLOAT, 64x1x5x5]
%onnx::Conv_849[FLOAT, 64x32x1x1]
%onnx::Conv_852[FLOAT, 384x64x1x1]
%onnx::Conv_853[FLOAT, 384]
%onnx::Conv_855[FLOAT, 384x1x5x5]
%onnx::Conv_858[FLOAT, 112x384x1x1]
%onnx::Conv_859[FLOAT, 112]
%onnx::Conv_861[FLOAT, 336x112x1x1]
%onnx::Conv_862[FLOAT, 336]
%onnx::Conv_864[FLOAT, 336x1x5x5]
%onnx::Conv_867[FLOAT, 112x336x1x1]
%onnx::Conv_870[FLOAT, 112x56x1x1]
%onnx::Conv_873[FLOAT, 112x1x5x5]
%onnx::Conv_876[FLOAT, 112x56x1x1]
%onnx::Conv_879[FLOAT, 112x56x1x1]
%onnx::Conv_882[FLOAT, 112x1x5x5]
%onnx::Conv_885[FLOAT, 112x56x1x1]
%onnx::Conv_888[FLOAT, 336x112x1x1]
%onnx::Conv_891[FLOAT, 336x1x5x5]
%onnx::Conv_894[FLOAT, 184x336x1x1]
%onnx::Conv_895[FLOAT, 184]
%onnx::Conv_897[FLOAT, 184x92x1x1]
%onnx::Conv_900[FLOAT, 184x1x5x5]
%onnx::Conv_903[FLOAT, 184x92x1x1]
%onnx::Conv_906[FLOAT, 552x184x1x1]
%onnx::Conv_907[FLOAT, 552]
%onnx::Conv_909[FLOAT, 552x1x3x3]
%onnx::Conv_912[FLOAT, 184x552x1x1]
%onnx::Conv_915[FLOAT, 1104x184x1x1]
%onnx::Conv_916[FLOAT, 1104]
%onnx::Conv_918[FLOAT, 1104x1x3x3]
%onnx::Conv_921[FLOAT, 184x1104x1x1]
%onnx::Conv_924[FLOAT, 552x184x1x1]
%onnx::Conv_927[FLOAT, 552x1x5x5]
%onnx::Conv_930[FLOAT, 352x552x1x1]
%onnx::Conv_931[FLOAT, 352]
%onnx::Conv_933[FLOAT, 1504x352x1x1]
%onnx::Conv_934[FLOAT, 1504]
) {
%onnx::Conv_928 = Identity(%onnx::Conv_907)
%onnx::Conv_925 = Identity(%onnx::Conv_907)
%onnx::Conv_922 = Identity(%onnx::Conv_895)
%onnx::Conv_919 = Identity(%onnx::Conv_916)
%onnx::Conv_913 = Identity(%onnx::Conv_895)
%onnx::Conv_910 = Identity(%onnx::Conv_907)
%onnx::Conv_904 = Identity(%onnx::Conv_895)
%onnx::Conv_901 = Identity(%onnx::Conv_895)
%onnx::Conv_898 = Identity(%onnx::Conv_895)
%onnx::Conv_892 = Identity(%onnx::Conv_862)
%onnx::Conv_889 = Identity(%onnx::Conv_862)
%onnx::Conv_886 = Identity(%onnx::Conv_859)
%onnx::Conv_883 = Identity(%onnx::Conv_859)
%onnx::Conv_880 = Identity(%onnx::Conv_859)
%onnx::Conv_877 = Identity(%onnx::Conv_859)
%onnx::Conv_874 = Identity(%onnx::Conv_859)
%onnx::Conv_871 = Identity(%onnx::Conv_859)
%onnx::Conv_868 = Identity(%onnx::Conv_859)
%onnx::Conv_865 = Identity(%onnx::Conv_862)
%onnx::Conv_856 = Identity(%onnx::Conv_853)
%onnx::Conv_850 = Identity(%onnx::Conv_823)
%onnx::Conv_847 = Identity(%onnx::Conv_823)
%onnx::Conv_844 = Identity(%onnx::Conv_823)
%onnx::Conv_841 = Identity(%onnx::Conv_823)
%onnx::Conv_838 = Identity(%onnx::Conv_823)
%onnx::Conv_835 = Identity(%onnx::Conv_823)
%onnx::Conv_832 = Identity(%onnx::Conv_823)
%onnx::Conv_829 = Identity(%onnx::Conv_790)
%onnx::Conv_826 = Identity(%onnx::Conv_790)
%onnx::Conv_820 = Identity(%onnx::Conv_787)
%onnx::Conv_817 = Identity(%onnx::Conv_787)
%onnx::Conv_814 = Identity(%onnx::Conv_787)
%onnx::Conv_811 = Identity(%onnx::Conv_787)
%onnx::Conv_808 = Identity(%onnx::Conv_787)
%onnx::Conv_805 = Identity(%onnx::Conv_787)
%onnx::Conv_802 = Identity(%onnx::Conv_787)
%onnx::Conv_799 = Identity(%onnx::Conv_787)
%onnx::Conv_796 = Identity(%onnx::Conv_787)
%onnx::Conv_793 = Identity(%onnx::Conv_790)
%onnx::Conv_784 = Identity(%onnx::Conv_760)
%onnx::Conv_781 = Identity(%onnx::Conv_760)
%onnx::Conv_778 = Identity(%onnx::Conv_760)
%onnx::Conv_775 = Identity(%onnx::Conv_760)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_760)
%onnx::Conv_763 = Identity(%onnx::Conv_760)
%onnx::Conv_757 = Identity(%onnx::Conv_742)
%onnx::Conv_754 = Identity(%onnx::Conv_742)
%onnx::Conv_751 = Identity(%onnx::Conv_742)
%onnx::Conv_748 = Identity(%onnx::Conv_745)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_741, %onnx::Conv_742)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_882, %onnx::Conv_883)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_885, %onnx::Conv_886)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_888, %onnx::Conv_889)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_891, %onnx::Conv_892)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_894, %onnx::Conv_895)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_897, %onnx::Conv_898)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_900, %onnx::Conv_901)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_903, %onnx::Conv_904)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_906, %onnx::Conv_907)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_909, %onnx::Conv_910)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_912, %onnx::Conv_913)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_915, %onnx::Conv_916)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_918, %onnx::Conv_919)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_921, %onnx::Conv_922)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_924, %onnx::Conv_925)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_927, %onnx::Conv_928)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_930, %onnx::Conv_931)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_933, %onnx::Conv_934)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%739 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %739
}
|
val_accuracy
| 0 | 56,808,576 | 2,048,644 |
{'zcp_synflow': 76.18746176199595, 'zcp_zen': 68.44309997558594, 'zcp_epe_nas': 15.590916298302991, 'zcp_fisher': 0.14862927794456482, 'zcp_flops': 56808576.0, 'zcp_grad_norm': 21.899150848388672, 'zcp_grasp': -0.08688735961914062, 'zcp_jacov': -16.058273312661502, 'zcp_l2_norm': 616.63525390625, 'zcp_nwot': 205.90953859056543, 'zcp_params': 2048644.0, 'zcp_plain': 0.005494223441928625, 'zcp_snip': 39.68156051635742, 'lat_1080ti_1': 0.7515884412959241, 'lat_1080ti_32': 0.5831218495704096, 'lat_1080ti_64': 0.33283795790455695, 'lat_2080ti_1': 0.7653820348518361, 'lat_2080ti_32': 0.5884980142841988, 'lat_2080ti_64': 0.36407434297806923, 'lat_essential_ph_1': 0.39622641509433965, 'lat_eyeriss': 0.31314983587919215, 'lat_fpga': 0.3355762475385665, 'lat_gold_6226': 0.3494969145718307, 'lat_gold_6240': 0.5967556889836344, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.3335383740703231, 'lat_raspi4': 0.4065723243830333, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.16535433070866143, 'lat_silver_4114': 0.638563231232414, 'lat_silver_4210r': 0.6568013139152766, 'lat_titan_rtx_1': 0.7372980205931654, 'lat_titan_rtx_32': 0.6037773772795495, 'lat_titan_rtx_64': 0.42914061217321103, 'lat_titanx_1': 0.3902073508953178, 'lat_titanx_32': 0.5018944681011585, 'lat_titanx_64': 0.3080425546778311, 'lat_titanxp_1': 0.7197805667584398, 'lat_titanxp_32': 0.5460891902548847, 'lat_titanxp_64': 0.3607439872593161}
| |
FBNet_2066
|
FBNet
|
2066
|
2066
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_642[FLOAT, 16x3x3x3]
%onnx::Conv_643[FLOAT, 16]
%onnx::Conv_645[FLOAT, 16x16x1x1]
%onnx::Conv_648[FLOAT, 16x1x5x5]
%onnx::Conv_651[FLOAT, 16x16x1x1]
%onnx::Conv_654[FLOAT, 16x8x1x1]
%onnx::Conv_657[FLOAT, 16x1x5x5]
%onnx::Conv_660[FLOAT, 24x8x1x1]
%onnx::Conv_661[FLOAT, 24]
%onnx::Conv_663[FLOAT, 24x24x1x1]
%onnx::Conv_666[FLOAT, 24x1x3x3]
%onnx::Conv_669[FLOAT, 24x24x1x1]
%onnx::Conv_672[FLOAT, 24x24x1x1]
%onnx::Conv_675[FLOAT, 24x1x3x3]
%onnx::Conv_678[FLOAT, 24x24x1x1]
%onnx::Conv_681[FLOAT, 24x24x1x1]
%onnx::Conv_684[FLOAT, 24x1x5x5]
%onnx::Conv_687[FLOAT, 24x24x1x1]
%onnx::Conv_690[FLOAT, 24x12x1x1]
%onnx::Conv_693[FLOAT, 24x1x3x3]
%onnx::Conv_696[FLOAT, 32x12x1x1]
%onnx::Conv_697[FLOAT, 32]
%onnx::Conv_699[FLOAT, 192x32x1x1]
%onnx::Conv_700[FLOAT, 192]
%onnx::Conv_702[FLOAT, 192x1x3x3]
%onnx::Conv_705[FLOAT, 32x192x1x1]
%onnx::Conv_708[FLOAT, 192x32x1x1]
%onnx::Conv_711[FLOAT, 192x1x3x3]
%onnx::Conv_714[FLOAT, 32x192x1x1]
%onnx::Conv_717[FLOAT, 192x32x1x1]
%onnx::Conv_720[FLOAT, 192x1x3x3]
%onnx::Conv_723[FLOAT, 64x192x1x1]
%onnx::Conv_724[FLOAT, 64]
%onnx::Conv_726[FLOAT, 64x32x1x1]
%onnx::Conv_729[FLOAT, 64x1x5x5]
%onnx::Conv_732[FLOAT, 64x32x1x1]
%onnx::Conv_735[FLOAT, 192x64x1x1]
%onnx::Conv_738[FLOAT, 192x1x3x3]
%onnx::Conv_741[FLOAT, 64x192x1x1]
%onnx::Conv_744[FLOAT, 64x64x1x1]
%onnx::Conv_747[FLOAT, 64x1x3x3]
%onnx::Conv_750[FLOAT, 64x64x1x1]
%onnx::Conv_753[FLOAT, 384x64x1x1]
%onnx::Conv_754[FLOAT, 384]
%onnx::Conv_756[FLOAT, 384x1x3x3]
%onnx::Conv_759[FLOAT, 112x384x1x1]
%onnx::Conv_760[FLOAT, 112]
%onnx::Conv_762[FLOAT, 336x112x1x1]
%onnx::Conv_763[FLOAT, 336]
%onnx::Conv_765[FLOAT, 336x1x5x5]
%onnx::Conv_768[FLOAT, 112x336x1x1]
%onnx::Conv_771[FLOAT, 112x112x1x1]
%onnx::Conv_774[FLOAT, 112x1x5x5]
%onnx::Conv_777[FLOAT, 112x112x1x1]
%onnx::Conv_780[FLOAT, 112x112x1x1]
%onnx::Conv_783[FLOAT, 112x1x3x3]
%onnx::Conv_786[FLOAT, 112x112x1x1]
%onnx::Conv_789[FLOAT, 184x112x1x1]
%onnx::Conv_790[FLOAT, 184]
%onnx::Conv_792[FLOAT, 1104x184x1x1]
%onnx::Conv_793[FLOAT, 1104]
%onnx::Conv_795[FLOAT, 1104x1x3x3]
%onnx::Conv_798[FLOAT, 184x1104x1x1]
%onnx::Conv_801[FLOAT, 184x92x1x1]
%onnx::Conv_804[FLOAT, 184x1x5x5]
%onnx::Conv_807[FLOAT, 184x92x1x1]
%onnx::Conv_810[FLOAT, 184x92x1x1]
%onnx::Conv_813[FLOAT, 184x1x3x3]
%onnx::Conv_816[FLOAT, 352x92x1x1]
%onnx::Conv_817[FLOAT, 352]
%onnx::Conv_819[FLOAT, 1504x352x1x1]
%onnx::Conv_820[FLOAT, 1504]
) {
%onnx::Conv_814 = Identity(%onnx::Conv_790)
%onnx::Conv_811 = Identity(%onnx::Conv_790)
%onnx::Conv_808 = Identity(%onnx::Conv_790)
%onnx::Conv_805 = Identity(%onnx::Conv_790)
%onnx::Conv_802 = Identity(%onnx::Conv_790)
%onnx::Conv_799 = Identity(%onnx::Conv_790)
%onnx::Conv_796 = Identity(%onnx::Conv_793)
%onnx::Conv_787 = Identity(%onnx::Conv_760)
%onnx::Conv_784 = Identity(%onnx::Conv_760)
%onnx::Conv_781 = Identity(%onnx::Conv_760)
%onnx::Conv_778 = Identity(%onnx::Conv_760)
%onnx::Conv_775 = Identity(%onnx::Conv_760)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_763)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_751 = Identity(%onnx::Conv_724)
%onnx::Conv_748 = Identity(%onnx::Conv_724)
%onnx::Conv_745 = Identity(%onnx::Conv_724)
%onnx::Conv_742 = Identity(%onnx::Conv_724)
%onnx::Conv_739 = Identity(%onnx::Conv_700)
%onnx::Conv_736 = Identity(%onnx::Conv_700)
%onnx::Conv_733 = Identity(%onnx::Conv_724)
%onnx::Conv_730 = Identity(%onnx::Conv_724)
%onnx::Conv_727 = Identity(%onnx::Conv_724)
%onnx::Conv_721 = Identity(%onnx::Conv_700)
%onnx::Conv_718 = Identity(%onnx::Conv_700)
%onnx::Conv_715 = Identity(%onnx::Conv_697)
%onnx::Conv_712 = Identity(%onnx::Conv_700)
%onnx::Conv_709 = Identity(%onnx::Conv_700)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_700)
%onnx::Conv_694 = Identity(%onnx::Conv_661)
%onnx::Conv_691 = Identity(%onnx::Conv_661)
%onnx::Conv_688 = Identity(%onnx::Conv_661)
%onnx::Conv_685 = Identity(%onnx::Conv_661)
%onnx::Conv_682 = Identity(%onnx::Conv_661)
%onnx::Conv_679 = Identity(%onnx::Conv_661)
%onnx::Conv_676 = Identity(%onnx::Conv_661)
%onnx::Conv_673 = Identity(%onnx::Conv_661)
%onnx::Conv_670 = Identity(%onnx::Conv_661)
%onnx::Conv_667 = Identity(%onnx::Conv_661)
%onnx::Conv_664 = Identity(%onnx::Conv_661)
%onnx::Conv_658 = Identity(%onnx::Conv_643)
%onnx::Conv_655 = Identity(%onnx::Conv_643)
%onnx::Conv_652 = Identity(%onnx::Conv_643)
%onnx::Conv_649 = Identity(%onnx::Conv_643)
%onnx::Conv_646 = Identity(%onnx::Conv_643)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_642, %onnx::Conv_643)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%640 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %640
}
|
val_accuracy
| 0 | 52,104,704 | 1,531,668 |
{'zcp_synflow': 72.56494337957142, 'zcp_zen': 62.65282440185547, 'zcp_epe_nas': 6.188530375966057, 'zcp_fisher': 0.11789791285991669, 'zcp_flops': 52104704.0, 'zcp_grad_norm': 20.730403900146484, 'zcp_grasp': -0.021427154541015625, 'zcp_jacov': -16.055589956048333, 'zcp_l2_norm': 553.8778686523438, 'zcp_nwot': 206.9422918393735, 'zcp_params': 1531668.0, 'zcp_plain': 0.006792387459427118, 'zcp_snip': 33.06220245361328, 'lat_1080ti_1': 0.593207961360602, 'lat_1080ti_32': 0.37823356765292804, 'lat_1080ti_64': 0.20505810902743998, 'lat_2080ti_1': 0.6487971408175138, 'lat_2080ti_32': 0.4036441270246552, 'lat_2080ti_64': 0.2378098296061592, 'lat_essential_ph_1': 0.22641509433962265, 'lat_eyeriss': 0.24312801422991553, 'lat_fpga': 0.28435864198141675, 'lat_gold_6226': 0.23203741102602785, 'lat_gold_6240': 0.4381643556860025, 'lat_pixel2': 0.15217391304347827, 'lat_pixel3': 0.2166802202852037, 'lat_raspi4': 0.24623597394337934, 'lat_samsung_a50': 0.10526315789473684, 'lat_samsung_s7': 0.11023622047244094, 'lat_silver_4114': 0.45905366585703994, 'lat_silver_4210r': 0.4770178562226773, 'lat_titan_rtx_1': 0.5269468850752093, 'lat_titan_rtx_32': 0.41572249137274236, 'lat_titan_rtx_64': 0.2714881447299462, 'lat_titanx_1': 0.27625628471172026, 'lat_titanx_32': 0.33093452891533975, 'lat_titanx_64': 0.2003749192527639, 'lat_titanxp_1': 0.516359538604654, 'lat_titanxp_32': 0.36960004403815344, 'lat_titanxp_64': 0.2347948544127191}
| |
FBNet_2399
|
FBNet
|
2399
|
2399
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_652[FLOAT, 16x3x3x3]
%onnx::Conv_653[FLOAT, 16]
%onnx::Conv_655[FLOAT, 16x16x1x1]
%onnx::Conv_658[FLOAT, 16x1x5x5]
%onnx::Conv_661[FLOAT, 16x16x1x1]
%onnx::Conv_664[FLOAT, 24x16x1x1]
%onnx::Conv_665[FLOAT, 24]
%onnx::Conv_667[FLOAT, 24x12x1x1]
%onnx::Conv_670[FLOAT, 24x1x5x5]
%onnx::Conv_673[FLOAT, 24x12x1x1]
%onnx::Conv_676[FLOAT, 144x24x1x1]
%onnx::Conv_677[FLOAT, 144]
%onnx::Conv_679[FLOAT, 144x1x3x3]
%onnx::Conv_682[FLOAT, 24x144x1x1]
%onnx::Conv_685[FLOAT, 24x12x1x1]
%onnx::Conv_688[FLOAT, 24x1x3x3]
%onnx::Conv_691[FLOAT, 24x12x1x1]
%onnx::Conv_694[FLOAT, 72x24x1x1]
%onnx::Conv_695[FLOAT, 72]
%onnx::Conv_697[FLOAT, 72x1x5x5]
%onnx::Conv_700[FLOAT, 32x72x1x1]
%onnx::Conv_701[FLOAT, 32]
%onnx::Conv_703[FLOAT, 96x32x1x1]
%onnx::Conv_704[FLOAT, 96]
%onnx::Conv_706[FLOAT, 96x1x3x3]
%onnx::Conv_709[FLOAT, 32x96x1x1]
%onnx::Conv_712[FLOAT, 192x32x1x1]
%onnx::Conv_713[FLOAT, 192]
%onnx::Conv_715[FLOAT, 192x1x3x3]
%onnx::Conv_718[FLOAT, 32x192x1x1]
%onnx::Conv_721[FLOAT, 96x32x1x1]
%onnx::Conv_724[FLOAT, 96x1x5x5]
%onnx::Conv_727[FLOAT, 64x96x1x1]
%onnx::Conv_728[FLOAT, 64]
%onnx::Conv_730[FLOAT, 64x32x1x1]
%onnx::Conv_733[FLOAT, 64x1x3x3]
%onnx::Conv_736[FLOAT, 64x32x1x1]
%onnx::Conv_739[FLOAT, 384x64x1x1]
%onnx::Conv_740[FLOAT, 384]
%onnx::Conv_742[FLOAT, 384x1x3x3]
%onnx::Conv_745[FLOAT, 64x384x1x1]
%onnx::Conv_748[FLOAT, 192x64x1x1]
%onnx::Conv_751[FLOAT, 192x1x3x3]
%onnx::Conv_754[FLOAT, 64x192x1x1]
%onnx::Conv_757[FLOAT, 384x64x1x1]
%onnx::Conv_760[FLOAT, 384x1x5x5]
%onnx::Conv_763[FLOAT, 112x384x1x1]
%onnx::Conv_764[FLOAT, 112]
%onnx::Conv_766[FLOAT, 112x112x1x1]
%onnx::Conv_769[FLOAT, 112x1x3x3]
%onnx::Conv_772[FLOAT, 112x112x1x1]
%onnx::Conv_775[FLOAT, 112x112x1x1]
%onnx::Conv_778[FLOAT, 112x1x3x3]
%onnx::Conv_781[FLOAT, 112x112x1x1]
%onnx::Conv_784[FLOAT, 112x112x1x1]
%onnx::Conv_787[FLOAT, 112x1x5x5]
%onnx::Conv_790[FLOAT, 112x112x1x1]
%onnx::Conv_793[FLOAT, 112x56x1x1]
%onnx::Conv_796[FLOAT, 112x1x3x3]
%onnx::Conv_799[FLOAT, 184x56x1x1]
%onnx::Conv_800[FLOAT, 184]
%onnx::Conv_802[FLOAT, 184x184x1x1]
%onnx::Conv_805[FLOAT, 184x1x5x5]
%onnx::Conv_808[FLOAT, 184x184x1x1]
%onnx::Conv_811[FLOAT, 184x184x1x1]
%onnx::Conv_814[FLOAT, 184x1x5x5]
%onnx::Conv_817[FLOAT, 184x184x1x1]
%onnx::Conv_820[FLOAT, 184x92x1x1]
%onnx::Conv_823[FLOAT, 184x1x5x5]
%onnx::Conv_826[FLOAT, 184x92x1x1]
%onnx::Conv_829[FLOAT, 352x184x1x1]
%onnx::Conv_830[FLOAT, 352]
%onnx::Conv_832[FLOAT, 1504x352x1x1]
%onnx::Conv_833[FLOAT, 1504]
) {
%onnx::Conv_827 = Identity(%onnx::Conv_800)
%onnx::Conv_824 = Identity(%onnx::Conv_800)
%onnx::Conv_821 = Identity(%onnx::Conv_800)
%onnx::Conv_818 = Identity(%onnx::Conv_800)
%onnx::Conv_815 = Identity(%onnx::Conv_800)
%onnx::Conv_812 = Identity(%onnx::Conv_800)
%onnx::Conv_809 = Identity(%onnx::Conv_800)
%onnx::Conv_806 = Identity(%onnx::Conv_800)
%onnx::Conv_803 = Identity(%onnx::Conv_800)
%onnx::Conv_797 = Identity(%onnx::Conv_764)
%onnx::Conv_794 = Identity(%onnx::Conv_764)
%onnx::Conv_791 = Identity(%onnx::Conv_764)
%onnx::Conv_788 = Identity(%onnx::Conv_764)
%onnx::Conv_785 = Identity(%onnx::Conv_764)
%onnx::Conv_782 = Identity(%onnx::Conv_764)
%onnx::Conv_779 = Identity(%onnx::Conv_764)
%onnx::Conv_776 = Identity(%onnx::Conv_764)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_764)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_761 = Identity(%onnx::Conv_740)
%onnx::Conv_758 = Identity(%onnx::Conv_740)
%onnx::Conv_755 = Identity(%onnx::Conv_728)
%onnx::Conv_752 = Identity(%onnx::Conv_713)
%onnx::Conv_749 = Identity(%onnx::Conv_713)
%onnx::Conv_746 = Identity(%onnx::Conv_728)
%onnx::Conv_743 = Identity(%onnx::Conv_740)
%onnx::Conv_737 = Identity(%onnx::Conv_728)
%onnx::Conv_734 = Identity(%onnx::Conv_728)
%onnx::Conv_731 = Identity(%onnx::Conv_728)
%onnx::Conv_725 = Identity(%onnx::Conv_704)
%onnx::Conv_722 = Identity(%onnx::Conv_704)
%onnx::Conv_719 = Identity(%onnx::Conv_701)
%onnx::Conv_716 = Identity(%onnx::Conv_713)
%onnx::Conv_710 = Identity(%onnx::Conv_701)
%onnx::Conv_707 = Identity(%onnx::Conv_704)
%onnx::Conv_698 = Identity(%onnx::Conv_695)
%onnx::Conv_692 = Identity(%onnx::Conv_665)
%onnx::Conv_689 = Identity(%onnx::Conv_665)
%onnx::Conv_686 = Identity(%onnx::Conv_665)
%onnx::Conv_683 = Identity(%onnx::Conv_665)
%onnx::Conv_680 = Identity(%onnx::Conv_677)
%onnx::Conv_674 = Identity(%onnx::Conv_665)
%onnx::Conv_671 = Identity(%onnx::Conv_665)
%onnx::Conv_668 = Identity(%onnx::Conv_665)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_652, %onnx::Conv_653)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.1/relu/Relu_output_0 = Relu(%/cells.1/conv/Conv_output_0)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/relu/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/relu/Relu_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%650 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %650
}
|
val_accuracy
| 0 | 53,090,432 | 1,254,356 |
{'zcp_synflow': 76.02088729821561, 'zcp_zen': 65.18929290771484, 'zcp_epe_nas': 18.385110975652488, 'zcp_fisher': 0.08393853157758713, 'zcp_flops': 53090432.0, 'zcp_grad_norm': 20.964176177978516, 'zcp_grasp': -0.040091514587402344, 'zcp_jacov': -16.056324337276187, 'zcp_l2_norm': 565.2555541992188, 'zcp_nwot': 211.08978415687167, 'zcp_params': 1254356.0, 'zcp_plain': -0.0008283951901830733, 'zcp_snip': 32.919132232666016, 'lat_1080ti_1': 0.5794112204657033, 'lat_1080ti_32': 0.5610402943319969, 'lat_1080ti_64': 0.3836192234737012, 'lat_2080ti_1': 0.5926863750100153, 'lat_2080ti_32': 0.5216643044709828, 'lat_2080ti_64': 0.39952218966011077, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.2788586727302735, 'lat_fpga': 0.25172424029527307, 'lat_gold_6226': 0.14643906885010838, 'lat_gold_6240': 0.34070570407591577, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.2573899232680736, 'lat_raspi4': 0.25599782774093643, 'lat_samsung_a50': 0.2, 'lat_samsung_s7': 0.10236220472440945, 'lat_silver_4114': 0.338744948144265, 'lat_silver_4210r': 0.3553384937819528, 'lat_titan_rtx_1': 0.5715871770644629, 'lat_titan_rtx_32': 0.5101393358850929, 'lat_titan_rtx_64': 0.4171081156759042, 'lat_titanx_1': 0.2873326769099003, 'lat_titanx_32': 0.4504459774327807, 'lat_titanx_64': 0.3612365293186477, 'lat_titanxp_1': 0.5174073453021298, 'lat_titanxp_32': 0.48632893688445733, 'lat_titanxp_64': 0.38816291582088375}
| |
FBNet_509
|
FBNet
|
509
|
509
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_643[FLOAT, 16x3x3x3]
%onnx::Conv_644[FLOAT, 16]
%onnx::Conv_646[FLOAT, 16x8x1x1]
%onnx::Conv_649[FLOAT, 16x1x5x5]
%onnx::Conv_652[FLOAT, 24x8x1x1]
%onnx::Conv_653[FLOAT, 24]
%onnx::Conv_655[FLOAT, 24x12x1x1]
%onnx::Conv_658[FLOAT, 24x1x5x5]
%onnx::Conv_661[FLOAT, 24x12x1x1]
%onnx::Conv_664[FLOAT, 144x24x1x1]
%onnx::Conv_665[FLOAT, 144]
%onnx::Conv_667[FLOAT, 144x1x3x3]
%onnx::Conv_670[FLOAT, 24x144x1x1]
%onnx::Conv_673[FLOAT, 72x24x1x1]
%onnx::Conv_674[FLOAT, 72]
%onnx::Conv_676[FLOAT, 72x1x3x3]
%onnx::Conv_679[FLOAT, 24x72x1x1]
%onnx::Conv_682[FLOAT, 144x24x1x1]
%onnx::Conv_685[FLOAT, 144x1x3x3]
%onnx::Conv_688[FLOAT, 32x144x1x1]
%onnx::Conv_689[FLOAT, 32]
%onnx::Conv_691[FLOAT, 96x32x1x1]
%onnx::Conv_692[FLOAT, 96]
%onnx::Conv_694[FLOAT, 96x1x5x5]
%onnx::Conv_697[FLOAT, 32x96x1x1]
%onnx::Conv_700[FLOAT, 96x32x1x1]
%onnx::Conv_703[FLOAT, 96x1x3x3]
%onnx::Conv_706[FLOAT, 32x96x1x1]
%onnx::Conv_709[FLOAT, 32x16x1x1]
%onnx::Conv_712[FLOAT, 32x1x3x3]
%onnx::Conv_715[FLOAT, 64x16x1x1]
%onnx::Conv_716[FLOAT, 64]
%onnx::Conv_718[FLOAT, 192x64x1x1]
%onnx::Conv_719[FLOAT, 192]
%onnx::Conv_721[FLOAT, 192x1x5x5]
%onnx::Conv_724[FLOAT, 64x192x1x1]
%onnx::Conv_727[FLOAT, 64x32x1x1]
%onnx::Conv_730[FLOAT, 64x1x3x3]
%onnx::Conv_733[FLOAT, 64x32x1x1]
%onnx::Conv_736[FLOAT, 192x64x1x1]
%onnx::Conv_739[FLOAT, 192x1x5x5]
%onnx::Conv_742[FLOAT, 64x192x1x1]
%onnx::Conv_745[FLOAT, 384x64x1x1]
%onnx::Conv_746[FLOAT, 384]
%onnx::Conv_748[FLOAT, 384x1x3x3]
%onnx::Conv_751[FLOAT, 112x384x1x1]
%onnx::Conv_752[FLOAT, 112]
%onnx::Conv_754[FLOAT, 112x56x1x1]
%onnx::Conv_757[FLOAT, 112x1x3x3]
%onnx::Conv_760[FLOAT, 112x56x1x1]
%onnx::Conv_763[FLOAT, 672x112x1x1]
%onnx::Conv_764[FLOAT, 672]
%onnx::Conv_766[FLOAT, 672x1x5x5]
%onnx::Conv_769[FLOAT, 112x672x1x1]
%onnx::Conv_772[FLOAT, 672x112x1x1]
%onnx::Conv_775[FLOAT, 672x1x3x3]
%onnx::Conv_778[FLOAT, 112x672x1x1]
%onnx::Conv_781[FLOAT, 184x112x1x1]
%onnx::Conv_782[FLOAT, 184]
%onnx::Conv_784[FLOAT, 1104x184x1x1]
%onnx::Conv_785[FLOAT, 1104]
%onnx::Conv_787[FLOAT, 1104x1x5x5]
%onnx::Conv_790[FLOAT, 184x1104x1x1]
%onnx::Conv_793[FLOAT, 1104x184x1x1]
%onnx::Conv_796[FLOAT, 1104x1x5x5]
%onnx::Conv_799[FLOAT, 184x1104x1x1]
%onnx::Conv_802[FLOAT, 184x184x1x1]
%onnx::Conv_805[FLOAT, 184x1x3x3]
%onnx::Conv_808[FLOAT, 184x184x1x1]
%onnx::Conv_811[FLOAT, 184x184x1x1]
%onnx::Conv_814[FLOAT, 184x1x5x5]
%onnx::Conv_817[FLOAT, 352x184x1x1]
%onnx::Conv_818[FLOAT, 352]
%onnx::Conv_820[FLOAT, 1504x352x1x1]
%onnx::Conv_821[FLOAT, 1504]
) {
%onnx::Conv_815 = Identity(%onnx::Conv_782)
%onnx::Conv_812 = Identity(%onnx::Conv_782)
%onnx::Conv_809 = Identity(%onnx::Conv_782)
%onnx::Conv_806 = Identity(%onnx::Conv_782)
%onnx::Conv_803 = Identity(%onnx::Conv_782)
%onnx::Conv_800 = Identity(%onnx::Conv_782)
%onnx::Conv_797 = Identity(%onnx::Conv_785)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_782)
%onnx::Conv_788 = Identity(%onnx::Conv_785)
%onnx::Conv_779 = Identity(%onnx::Conv_752)
%onnx::Conv_776 = Identity(%onnx::Conv_764)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_752)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_761 = Identity(%onnx::Conv_752)
%onnx::Conv_758 = Identity(%onnx::Conv_752)
%onnx::Conv_755 = Identity(%onnx::Conv_752)
%onnx::Conv_749 = Identity(%onnx::Conv_746)
%onnx::Conv_743 = Identity(%onnx::Conv_716)
%onnx::Conv_740 = Identity(%onnx::Conv_719)
%onnx::Conv_737 = Identity(%onnx::Conv_719)
%onnx::Conv_734 = Identity(%onnx::Conv_716)
%onnx::Conv_731 = Identity(%onnx::Conv_716)
%onnx::Conv_728 = Identity(%onnx::Conv_716)
%onnx::Conv_725 = Identity(%onnx::Conv_716)
%onnx::Conv_722 = Identity(%onnx::Conv_719)
%onnx::Conv_713 = Identity(%onnx::Conv_689)
%onnx::Conv_710 = Identity(%onnx::Conv_689)
%onnx::Conv_707 = Identity(%onnx::Conv_689)
%onnx::Conv_704 = Identity(%onnx::Conv_692)
%onnx::Conv_701 = Identity(%onnx::Conv_692)
%onnx::Conv_698 = Identity(%onnx::Conv_689)
%onnx::Conv_695 = Identity(%onnx::Conv_692)
%onnx::Conv_686 = Identity(%onnx::Conv_665)
%onnx::Conv_683 = Identity(%onnx::Conv_665)
%onnx::Conv_680 = Identity(%onnx::Conv_653)
%onnx::Conv_677 = Identity(%onnx::Conv_674)
%onnx::Conv_671 = Identity(%onnx::Conv_653)
%onnx::Conv_668 = Identity(%onnx::Conv_665)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_650 = Identity(%onnx::Conv_644)
%onnx::Conv_647 = Identity(%onnx::Conv_644)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_643, %onnx::Conv_644)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%641 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %641
}
|
val_accuracy
| 0 | 83,478,784 | 2,283,108 |
{'zcp_synflow': 74.2818409487285, 'zcp_zen': 68.38712310791016, 'zcp_epe_nas': 34.92888830274783, 'zcp_fisher': 0.06951306760311127, 'zcp_flops': 83478784.0, 'zcp_grad_norm': 20.416166305541992, 'zcp_grasp': -0.006525993347167969, 'zcp_jacov': -16.06401727677944, 'zcp_l2_norm': 660.7625732421875, 'zcp_nwot': 214.69018139192278, 'zcp_params': 2283108.0, 'zcp_plain': 0.003367349971085787, 'zcp_snip': 39.05625915527344, 'lat_1080ti_1': 0.6073970705190774, 'lat_1080ti_32': 0.47854784217413127, 'lat_1080ti_64': 0.4822220825248549, 'lat_2080ti_1': 0.5418703027974832, 'lat_2080ti_32': 0.524414960374097, 'lat_2080ti_64': 0.49688361007612825, 'lat_essential_ph_1': 0.4339622641509434, 'lat_eyeriss': 0.583123979975428, 'lat_fpga': 0.6930308037878862, 'lat_gold_6226': 0.5011012976365681, 'lat_gold_6240': 0.630431029715175, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.5697942268790509, 'lat_raspi4': 0.6090164522432543, 'lat_samsung_a50': 0.2631578947368421, 'lat_samsung_s7': 0.29133858267716534, 'lat_silver_4114': 0.6331249759200807, 'lat_silver_4210r': 0.6238863137586628, 'lat_titan_rtx_1': 0.5219495738522212, 'lat_titan_rtx_32': 0.49338855044270385, 'lat_titan_rtx_64': 0.48990528570085445, 'lat_titanx_1': 0.2856764093380064, 'lat_titanx_32': 0.49566279922335404, 'lat_titanx_64': 0.4786086217622535, 'lat_titanxp_1': 0.5034313830689543, 'lat_titanxp_32': 0.47878234593791297, 'lat_titanxp_64': 0.4831644921914947}
| |
FBNet_4315
|
FBNet
|
4315
|
4315
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_650[FLOAT, 16x3x3x3]
%onnx::Conv_651[FLOAT, 16]
%onnx::Conv_653[FLOAT, 48x16x1x1]
%onnx::Conv_654[FLOAT, 48]
%onnx::Conv_656[FLOAT, 48x1x5x5]
%onnx::Conv_659[FLOAT, 16x48x1x1]
%onnx::Conv_662[FLOAT, 48x16x1x1]
%onnx::Conv_665[FLOAT, 48x1x3x3]
%onnx::Conv_668[FLOAT, 24x48x1x1]
%onnx::Conv_669[FLOAT, 24]
%onnx::Conv_671[FLOAT, 24x24x1x1]
%onnx::Conv_674[FLOAT, 24x1x5x5]
%onnx::Conv_677[FLOAT, 24x24x1x1]
%onnx::Conv_680[FLOAT, 144x24x1x1]
%onnx::Conv_681[FLOAT, 144]
%onnx::Conv_683[FLOAT, 144x1x5x5]
%onnx::Conv_686[FLOAT, 24x144x1x1]
%onnx::Conv_689[FLOAT, 24x12x1x1]
%onnx::Conv_692[FLOAT, 24x1x5x5]
%onnx::Conv_695[FLOAT, 32x12x1x1]
%onnx::Conv_696[FLOAT, 32]
%onnx::Conv_698[FLOAT, 32x32x1x1]
%onnx::Conv_701[FLOAT, 32x1x5x5]
%onnx::Conv_704[FLOAT, 32x32x1x1]
%onnx::Conv_707[FLOAT, 32x32x1x1]
%onnx::Conv_710[FLOAT, 32x1x3x3]
%onnx::Conv_713[FLOAT, 32x32x1x1]
%onnx::Conv_716[FLOAT, 32x16x1x1]
%onnx::Conv_719[FLOAT, 32x1x5x5]
%onnx::Conv_722[FLOAT, 32x16x1x1]
%onnx::Conv_725[FLOAT, 32x16x1x1]
%onnx::Conv_728[FLOAT, 32x1x3x3]
%onnx::Conv_731[FLOAT, 64x16x1x1]
%onnx::Conv_732[FLOAT, 64]
%onnx::Conv_734[FLOAT, 64x32x1x1]
%onnx::Conv_737[FLOAT, 64x1x3x3]
%onnx::Conv_740[FLOAT, 64x32x1x1]
%onnx::Conv_743[FLOAT, 64x32x1x1]
%onnx::Conv_746[FLOAT, 64x1x3x3]
%onnx::Conv_749[FLOAT, 64x32x1x1]
%onnx::Conv_752[FLOAT, 192x64x1x1]
%onnx::Conv_753[FLOAT, 192]
%onnx::Conv_755[FLOAT, 192x1x3x3]
%onnx::Conv_758[FLOAT, 64x192x1x1]
%onnx::Conv_761[FLOAT, 64x64x1x1]
%onnx::Conv_764[FLOAT, 64x1x5x5]
%onnx::Conv_767[FLOAT, 112x64x1x1]
%onnx::Conv_768[FLOAT, 112]
%onnx::Conv_770[FLOAT, 112x112x1x1]
%onnx::Conv_773[FLOAT, 112x1x5x5]
%onnx::Conv_776[FLOAT, 112x112x1x1]
%onnx::Conv_779[FLOAT, 336x112x1x1]
%onnx::Conv_780[FLOAT, 336]
%onnx::Conv_782[FLOAT, 336x1x5x5]
%onnx::Conv_785[FLOAT, 184x336x1x1]
%onnx::Conv_786[FLOAT, 184]
%onnx::Conv_788[FLOAT, 184x92x1x1]
%onnx::Conv_791[FLOAT, 184x1x3x3]
%onnx::Conv_794[FLOAT, 184x92x1x1]
%onnx::Conv_797[FLOAT, 1104x184x1x1]
%onnx::Conv_798[FLOAT, 1104]
%onnx::Conv_800[FLOAT, 1104x1x5x5]
%onnx::Conv_803[FLOAT, 184x1104x1x1]
%onnx::Conv_806[FLOAT, 184x184x1x1]
%onnx::Conv_809[FLOAT, 184x1x3x3]
%onnx::Conv_812[FLOAT, 184x184x1x1]
%onnx::Conv_815[FLOAT, 184x184x1x1]
%onnx::Conv_818[FLOAT, 184x1x3x3]
%onnx::Conv_821[FLOAT, 352x184x1x1]
%onnx::Conv_822[FLOAT, 352]
%onnx::Conv_824[FLOAT, 1504x352x1x1]
%onnx::Conv_825[FLOAT, 1504]
) {
%onnx::Conv_819 = Identity(%onnx::Conv_786)
%onnx::Conv_816 = Identity(%onnx::Conv_786)
%onnx::Conv_813 = Identity(%onnx::Conv_786)
%onnx::Conv_810 = Identity(%onnx::Conv_786)
%onnx::Conv_807 = Identity(%onnx::Conv_786)
%onnx::Conv_804 = Identity(%onnx::Conv_786)
%onnx::Conv_801 = Identity(%onnx::Conv_798)
%onnx::Conv_795 = Identity(%onnx::Conv_786)
%onnx::Conv_792 = Identity(%onnx::Conv_786)
%onnx::Conv_789 = Identity(%onnx::Conv_786)
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_768)
%onnx::Conv_771 = Identity(%onnx::Conv_768)
%onnx::Conv_765 = Identity(%onnx::Conv_732)
%onnx::Conv_762 = Identity(%onnx::Conv_732)
%onnx::Conv_759 = Identity(%onnx::Conv_732)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_750 = Identity(%onnx::Conv_732)
%onnx::Conv_747 = Identity(%onnx::Conv_732)
%onnx::Conv_744 = Identity(%onnx::Conv_732)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_732)
%onnx::Conv_735 = Identity(%onnx::Conv_732)
%onnx::Conv_729 = Identity(%onnx::Conv_696)
%onnx::Conv_726 = Identity(%onnx::Conv_696)
%onnx::Conv_723 = Identity(%onnx::Conv_696)
%onnx::Conv_720 = Identity(%onnx::Conv_696)
%onnx::Conv_717 = Identity(%onnx::Conv_696)
%onnx::Conv_714 = Identity(%onnx::Conv_696)
%onnx::Conv_711 = Identity(%onnx::Conv_696)
%onnx::Conv_708 = Identity(%onnx::Conv_696)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_696)
%onnx::Conv_699 = Identity(%onnx::Conv_696)
%onnx::Conv_693 = Identity(%onnx::Conv_669)
%onnx::Conv_690 = Identity(%onnx::Conv_669)
%onnx::Conv_687 = Identity(%onnx::Conv_669)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_678 = Identity(%onnx::Conv_669)
%onnx::Conv_675 = Identity(%onnx::Conv_669)
%onnx::Conv_672 = Identity(%onnx::Conv_669)
%onnx::Conv_666 = Identity(%onnx::Conv_654)
%onnx::Conv_663 = Identity(%onnx::Conv_654)
%onnx::Conv_660 = Identity(%onnx::Conv_651)
%onnx::Conv_657 = Identity(%onnx::Conv_654)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_650, %onnx::Conv_651)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%648 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %648
}
|
val_accuracy
| 0 | 50,436,224 | 1,549,308 |
{'zcp_synflow': 70.99698008682198, 'zcp_zen': 60.3331184387207, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.12822341918945312, 'zcp_flops': 50436224.0, 'zcp_grad_norm': 19.243751525878906, 'zcp_grasp': 0.07659912109375, 'zcp_jacov': -16.058558687297396, 'zcp_l2_norm': 516.66943359375, 'zcp_nwot': 208.33009284709482, 'zcp_params': 1549308.0, 'zcp_plain': 0.0019968266133219004, 'zcp_snip': 31.469764709472656, 'lat_1080ti_1': 0.5639733099618793, 'lat_1080ti_32': 0.49842317636940303, 'lat_1080ti_64': 0.41513177466026646, 'lat_2080ti_1': 0.5568567608375531, 'lat_2080ti_32': 0.5302232696168474, 'lat_2080ti_64': 0.4295245446345071, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.27201877762088994, 'lat_fpga': 0.21204444928210256, 'lat_gold_6226': 0.1687134210226999, 'lat_gold_6240': 0.3046135328904421, 'lat_pixel2': 0.15217391304347827, 'lat_pixel3': 0.2992436810048828, 'lat_raspi4': 0.3142134150473346, 'lat_samsung_a50': 0.11578947368421053, 'lat_samsung_s7': 0.05511811023622047, 'lat_silver_4114': 0.324761170135245, 'lat_silver_4210r': 0.3739539473288171, 'lat_titan_rtx_1': 0.509926992018158, 'lat_titan_rtx_32': 0.5203206000315591, 'lat_titan_rtx_64': 0.4502662457705475, 'lat_titanx_1': 0.2702044748817578, 'lat_titanx_32': 0.4787583649525663, 'lat_titanx_64': 0.3951147201965231, 'lat_titanxp_1': 0.48646022434359487, 'lat_titanxp_32': 0.5015700499439107, 'lat_titanxp_64': 0.45020006970435367}
| |
FBNet_2056
|
FBNet
|
2056
|
2056
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_586[FLOAT, 16x3x3x3]
%onnx::Conv_587[FLOAT, 16]
%onnx::Conv_589[FLOAT, 16x16x1x1]
%onnx::Conv_592[FLOAT, 16x1x3x3]
%onnx::Conv_595[FLOAT, 16x16x1x1]
%onnx::Conv_598[FLOAT, 16x8x1x1]
%onnx::Conv_601[FLOAT, 16x1x3x3]
%onnx::Conv_604[FLOAT, 24x8x1x1]
%onnx::Conv_605[FLOAT, 24]
%onnx::Conv_607[FLOAT, 72x24x1x1]
%onnx::Conv_608[FLOAT, 72]
%onnx::Conv_610[FLOAT, 72x1x3x3]
%onnx::Conv_613[FLOAT, 24x72x1x1]
%onnx::Conv_616[FLOAT, 144x24x1x1]
%onnx::Conv_617[FLOAT, 144]
%onnx::Conv_619[FLOAT, 144x1x5x5]
%onnx::Conv_622[FLOAT, 24x144x1x1]
%onnx::Conv_625[FLOAT, 24x24x1x1]
%onnx::Conv_628[FLOAT, 24x1x3x3]
%onnx::Conv_631[FLOAT, 32x24x1x1]
%onnx::Conv_632[FLOAT, 32]
%onnx::Conv_634[FLOAT, 32x32x1x1]
%onnx::Conv_637[FLOAT, 32x1x5x5]
%onnx::Conv_640[FLOAT, 32x32x1x1]
%onnx::Conv_643[FLOAT, 192x32x1x1]
%onnx::Conv_644[FLOAT, 192]
%onnx::Conv_646[FLOAT, 192x1x3x3]
%onnx::Conv_649[FLOAT, 32x192x1x1]
%onnx::Conv_652[FLOAT, 32x32x1x1]
%onnx::Conv_655[FLOAT, 32x1x3x3]
%onnx::Conv_658[FLOAT, 32x32x1x1]
%onnx::Conv_661[FLOAT, 64x32x1x1]
%onnx::Conv_662[FLOAT, 64]
%onnx::Conv_664[FLOAT, 192x64x1x1]
%onnx::Conv_667[FLOAT, 192x1x3x3]
%onnx::Conv_670[FLOAT, 64x192x1x1]
%onnx::Conv_673[FLOAT, 192x64x1x1]
%onnx::Conv_676[FLOAT, 192x1x3x3]
%onnx::Conv_679[FLOAT, 64x192x1x1]
%onnx::Conv_682[FLOAT, 192x64x1x1]
%onnx::Conv_685[FLOAT, 192x1x5x5]
%onnx::Conv_688[FLOAT, 64x192x1x1]
%onnx::Conv_691[FLOAT, 64x64x1x1]
%onnx::Conv_694[FLOAT, 64x1x5x5]
%onnx::Conv_697[FLOAT, 112x64x1x1]
%onnx::Conv_698[FLOAT, 112]
%onnx::Conv_700[FLOAT, 672x112x1x1]
%onnx::Conv_701[FLOAT, 672]
%onnx::Conv_703[FLOAT, 672x1x3x3]
%onnx::Conv_706[FLOAT, 112x672x1x1]
%onnx::Conv_709[FLOAT, 336x112x1x1]
%onnx::Conv_710[FLOAT, 336]
%onnx::Conv_712[FLOAT, 336x1x5x5]
%onnx::Conv_715[FLOAT, 112x336x1x1]
%onnx::Conv_718[FLOAT, 672x112x1x1]
%onnx::Conv_721[FLOAT, 672x1x5x5]
%onnx::Conv_724[FLOAT, 112x672x1x1]
%onnx::Conv_727[FLOAT, 112x56x1x1]
%onnx::Conv_730[FLOAT, 112x1x3x3]
%onnx::Conv_733[FLOAT, 184x56x1x1]
%onnx::Conv_734[FLOAT, 184]
%onnx::Conv_736[FLOAT, 552x184x1x1]
%onnx::Conv_737[FLOAT, 552]
%onnx::Conv_739[FLOAT, 552x1x3x3]
%onnx::Conv_742[FLOAT, 184x552x1x1]
%onnx::Conv_745[FLOAT, 1104x184x1x1]
%onnx::Conv_746[FLOAT, 1104]
%onnx::Conv_748[FLOAT, 1104x1x5x5]
%onnx::Conv_751[FLOAT, 184x1104x1x1]
%onnx::Conv_754[FLOAT, 1104x184x1x1]
%onnx::Conv_757[FLOAT, 1104x1x3x3]
%onnx::Conv_760[FLOAT, 352x1104x1x1]
%onnx::Conv_761[FLOAT, 352]
%onnx::Conv_763[FLOAT, 1504x352x1x1]
%onnx::Conv_764[FLOAT, 1504]
) {
%onnx::Conv_758 = Identity(%onnx::Conv_746)
%onnx::Conv_755 = Identity(%onnx::Conv_746)
%onnx::Conv_752 = Identity(%onnx::Conv_734)
%onnx::Conv_749 = Identity(%onnx::Conv_746)
%onnx::Conv_743 = Identity(%onnx::Conv_734)
%onnx::Conv_740 = Identity(%onnx::Conv_737)
%onnx::Conv_731 = Identity(%onnx::Conv_698)
%onnx::Conv_728 = Identity(%onnx::Conv_698)
%onnx::Conv_725 = Identity(%onnx::Conv_698)
%onnx::Conv_722 = Identity(%onnx::Conv_701)
%onnx::Conv_719 = Identity(%onnx::Conv_701)
%onnx::Conv_716 = Identity(%onnx::Conv_698)
%onnx::Conv_713 = Identity(%onnx::Conv_710)
%onnx::Conv_707 = Identity(%onnx::Conv_698)
%onnx::Conv_704 = Identity(%onnx::Conv_701)
%onnx::Conv_695 = Identity(%onnx::Conv_662)
%onnx::Conv_692 = Identity(%onnx::Conv_662)
%onnx::Conv_689 = Identity(%onnx::Conv_662)
%onnx::Conv_686 = Identity(%onnx::Conv_644)
%onnx::Conv_683 = Identity(%onnx::Conv_644)
%onnx::Conv_680 = Identity(%onnx::Conv_662)
%onnx::Conv_677 = Identity(%onnx::Conv_644)
%onnx::Conv_674 = Identity(%onnx::Conv_644)
%onnx::Conv_671 = Identity(%onnx::Conv_662)
%onnx::Conv_668 = Identity(%onnx::Conv_644)
%onnx::Conv_665 = Identity(%onnx::Conv_644)
%onnx::Conv_659 = Identity(%onnx::Conv_632)
%onnx::Conv_656 = Identity(%onnx::Conv_632)
%onnx::Conv_653 = Identity(%onnx::Conv_632)
%onnx::Conv_650 = Identity(%onnx::Conv_632)
%onnx::Conv_647 = Identity(%onnx::Conv_644)
%onnx::Conv_641 = Identity(%onnx::Conv_632)
%onnx::Conv_638 = Identity(%onnx::Conv_632)
%onnx::Conv_635 = Identity(%onnx::Conv_632)
%onnx::Conv_629 = Identity(%onnx::Conv_605)
%onnx::Conv_626 = Identity(%onnx::Conv_605)
%onnx::Conv_623 = Identity(%onnx::Conv_605)
%onnx::Conv_620 = Identity(%onnx::Conv_617)
%onnx::Conv_614 = Identity(%onnx::Conv_605)
%onnx::Conv_611 = Identity(%onnx::Conv_608)
%onnx::Conv_602 = Identity(%onnx::Conv_587)
%onnx::Conv_599 = Identity(%onnx::Conv_587)
%onnx::Conv_596 = Identity(%onnx::Conv_587)
%onnx::Conv_593 = Identity(%onnx::Conv_587)
%onnx::Conv_590 = Identity(%onnx::Conv_587)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_586, %onnx::Conv_587)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_589, %onnx::Conv_590)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_592, %onnx::Conv_593)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_595, %onnx::Conv_596)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_598, %onnx::Conv_599)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_601, %onnx::Conv_602)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_604, %onnx::Conv_605)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_607, %onnx::Conv_608)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_610, %onnx::Conv_611)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_613, %onnx::Conv_614)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%584 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %584
}
|
val_accuracy
| 0 | 86,862,720 | 2,512,108 |
{'zcp_synflow': 76.5100042929323, 'zcp_zen': 67.44517517089844, 'zcp_epe_nas': 10.206705849777528, 'zcp_fisher': 0.11529095470905304, 'zcp_flops': 86862720.0, 'zcp_grad_norm': 22.90542221069336, 'zcp_grasp': -0.042835235595703125, 'zcp_jacov': -16.06546000840636, 'zcp_l2_norm': 666.1715087890625, 'zcp_nwot': 213.12093395329373, 'zcp_params': 2512108.0, 'zcp_plain': -0.0029155102092772722, 'zcp_snip': 38.64791488647461, 'lat_1080ti_1': 0.4817481426951119, 'lat_1080ti_32': 0.4923498238627656, 'lat_1080ti_64': 0.4252952940906183, 'lat_2080ti_1': 0.5101011447869717, 'lat_2080ti_32': 0.5038483166962726, 'lat_2080ti_64': 0.4614751718023518, 'lat_essential_ph_1': 0.41509433962264153, 'lat_eyeriss': 0.5994535418920649, 'lat_fpga': 0.7542623615907537, 'lat_gold_6226': 0.5497878094416546, 'lat_gold_6240': 0.5946893068170002, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.6091242014802866, 'lat_raspi4': 0.664667921301059, 'lat_samsung_a50': 0.25263157894736843, 'lat_samsung_s7': 0.2283464566929134, 'lat_silver_4114': 0.627577294438341, 'lat_silver_4210r': 0.5782618138225859, 'lat_titan_rtx_1': 0.4713707402797177, 'lat_titan_rtx_32': 0.4825274867632265, 'lat_titan_rtx_64': 0.45643428999386354, 'lat_titanx_1': 0.2533254454895204, 'lat_titanx_32': 0.4629065630314837, 'lat_titanx_64': 0.48249793847555555, 'lat_titanxp_1': 0.451580888576864, 'lat_titanxp_32': 0.4831562290584534, 'lat_titanxp_64': 0.4430731433300539}
| |
FBNet_4520
|
FBNet
|
4520
|
4520
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_632[FLOAT, 16x3x3x3]
%onnx::Conv_633[FLOAT, 16]
%onnx::Conv_635[FLOAT, 48x16x1x1]
%onnx::Conv_636[FLOAT, 48]
%onnx::Conv_638[FLOAT, 48x1x3x3]
%onnx::Conv_641[FLOAT, 16x48x1x1]
%onnx::Conv_644[FLOAT, 96x16x1x1]
%onnx::Conv_645[FLOAT, 96]
%onnx::Conv_647[FLOAT, 96x1x3x3]
%onnx::Conv_650[FLOAT, 24x96x1x1]
%onnx::Conv_651[FLOAT, 24]
%onnx::Conv_653[FLOAT, 24x12x1x1]
%onnx::Conv_656[FLOAT, 24x1x5x5]
%onnx::Conv_659[FLOAT, 24x12x1x1]
%onnx::Conv_662[FLOAT, 144x24x1x1]
%onnx::Conv_663[FLOAT, 144]
%onnx::Conv_665[FLOAT, 144x1x3x3]
%onnx::Conv_668[FLOAT, 24x144x1x1]
%onnx::Conv_671[FLOAT, 144x24x1x1]
%onnx::Conv_674[FLOAT, 144x1x3x3]
%onnx::Conv_677[FLOAT, 32x144x1x1]
%onnx::Conv_678[FLOAT, 32]
%onnx::Conv_680[FLOAT, 32x16x1x1]
%onnx::Conv_683[FLOAT, 32x1x3x3]
%onnx::Conv_686[FLOAT, 32x16x1x1]
%onnx::Conv_689[FLOAT, 32x32x1x1]
%onnx::Conv_692[FLOAT, 32x1x3x3]
%onnx::Conv_695[FLOAT, 32x32x1x1]
%onnx::Conv_698[FLOAT, 96x32x1x1]
%onnx::Conv_701[FLOAT, 96x1x3x3]
%onnx::Conv_704[FLOAT, 32x96x1x1]
%onnx::Conv_707[FLOAT, 96x32x1x1]
%onnx::Conv_710[FLOAT, 96x1x5x5]
%onnx::Conv_713[FLOAT, 64x96x1x1]
%onnx::Conv_714[FLOAT, 64]
%onnx::Conv_716[FLOAT, 64x64x1x1]
%onnx::Conv_719[FLOAT, 64x1x5x5]
%onnx::Conv_722[FLOAT, 64x64x1x1]
%onnx::Conv_725[FLOAT, 64x64x1x1]
%onnx::Conv_728[FLOAT, 64x1x5x5]
%onnx::Conv_731[FLOAT, 64x64x1x1]
%onnx::Conv_734[FLOAT, 64x64x1x1]
%onnx::Conv_737[FLOAT, 64x1x3x3]
%onnx::Conv_740[FLOAT, 64x64x1x1]
%onnx::Conv_743[FLOAT, 112x64x1x1]
%onnx::Conv_744[FLOAT, 112]
%onnx::Conv_746[FLOAT, 336x112x1x1]
%onnx::Conv_747[FLOAT, 336]
%onnx::Conv_749[FLOAT, 336x1x5x5]
%onnx::Conv_752[FLOAT, 112x336x1x1]
%onnx::Conv_755[FLOAT, 336x112x1x1]
%onnx::Conv_758[FLOAT, 336x1x3x3]
%onnx::Conv_761[FLOAT, 112x336x1x1]
%onnx::Conv_764[FLOAT, 336x112x1x1]
%onnx::Conv_767[FLOAT, 336x1x5x5]
%onnx::Conv_770[FLOAT, 112x336x1x1]
%onnx::Conv_773[FLOAT, 672x112x1x1]
%onnx::Conv_774[FLOAT, 672]
%onnx::Conv_776[FLOAT, 672x1x3x3]
%onnx::Conv_779[FLOAT, 184x672x1x1]
%onnx::Conv_780[FLOAT, 184]
%onnx::Conv_782[FLOAT, 184x92x1x1]
%onnx::Conv_785[FLOAT, 184x1x3x3]
%onnx::Conv_788[FLOAT, 184x92x1x1]
%onnx::Conv_791[FLOAT, 184x184x1x1]
%onnx::Conv_794[FLOAT, 184x1x5x5]
%onnx::Conv_797[FLOAT, 184x184x1x1]
%onnx::Conv_800[FLOAT, 184x184x1x1]
%onnx::Conv_803[FLOAT, 184x1x5x5]
%onnx::Conv_806[FLOAT, 184x184x1x1]
%onnx::Conv_809[FLOAT, 184x184x1x1]
%onnx::Conv_812[FLOAT, 184x1x3x3]
%onnx::Conv_815[FLOAT, 352x184x1x1]
%onnx::Conv_816[FLOAT, 352]
%onnx::Conv_818[FLOAT, 1504x352x1x1]
%onnx::Conv_819[FLOAT, 1504]
) {
%onnx::Conv_813 = Identity(%onnx::Conv_780)
%onnx::Conv_810 = Identity(%onnx::Conv_780)
%onnx::Conv_807 = Identity(%onnx::Conv_780)
%onnx::Conv_804 = Identity(%onnx::Conv_780)
%onnx::Conv_801 = Identity(%onnx::Conv_780)
%onnx::Conv_798 = Identity(%onnx::Conv_780)
%onnx::Conv_795 = Identity(%onnx::Conv_780)
%onnx::Conv_792 = Identity(%onnx::Conv_780)
%onnx::Conv_789 = Identity(%onnx::Conv_780)
%onnx::Conv_786 = Identity(%onnx::Conv_780)
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_777 = Identity(%onnx::Conv_774)
%onnx::Conv_771 = Identity(%onnx::Conv_744)
%onnx::Conv_768 = Identity(%onnx::Conv_747)
%onnx::Conv_765 = Identity(%onnx::Conv_747)
%onnx::Conv_762 = Identity(%onnx::Conv_744)
%onnx::Conv_759 = Identity(%onnx::Conv_747)
%onnx::Conv_756 = Identity(%onnx::Conv_747)
%onnx::Conv_753 = Identity(%onnx::Conv_744)
%onnx::Conv_750 = Identity(%onnx::Conv_747)
%onnx::Conv_741 = Identity(%onnx::Conv_714)
%onnx::Conv_738 = Identity(%onnx::Conv_714)
%onnx::Conv_735 = Identity(%onnx::Conv_714)
%onnx::Conv_732 = Identity(%onnx::Conv_714)
%onnx::Conv_729 = Identity(%onnx::Conv_714)
%onnx::Conv_726 = Identity(%onnx::Conv_714)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_714)
%onnx::Conv_717 = Identity(%onnx::Conv_714)
%onnx::Conv_711 = Identity(%onnx::Conv_645)
%onnx::Conv_708 = Identity(%onnx::Conv_645)
%onnx::Conv_705 = Identity(%onnx::Conv_678)
%onnx::Conv_702 = Identity(%onnx::Conv_645)
%onnx::Conv_699 = Identity(%onnx::Conv_645)
%onnx::Conv_696 = Identity(%onnx::Conv_678)
%onnx::Conv_693 = Identity(%onnx::Conv_678)
%onnx::Conv_690 = Identity(%onnx::Conv_678)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_678)
%onnx::Conv_681 = Identity(%onnx::Conv_678)
%onnx::Conv_675 = Identity(%onnx::Conv_663)
%onnx::Conv_672 = Identity(%onnx::Conv_663)
%onnx::Conv_669 = Identity(%onnx::Conv_651)
%onnx::Conv_666 = Identity(%onnx::Conv_663)
%onnx::Conv_660 = Identity(%onnx::Conv_651)
%onnx::Conv_657 = Identity(%onnx::Conv_651)
%onnx::Conv_654 = Identity(%onnx::Conv_651)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%onnx::Conv_642 = Identity(%onnx::Conv_633)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_632, %onnx::Conv_633)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%630 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %630
}
|
val_accuracy
| 0 | 67,347,328 | 1,514,876 |
{'zcp_synflow': 79.43717686684118, 'zcp_zen': 68.62718200683594, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.17175163328647614, 'zcp_flops': 67347328.0, 'zcp_grad_norm': 24.019357681274414, 'zcp_grasp': 0.06924057006835938, 'zcp_jacov': -16.057957424625712, 'zcp_l2_norm': 616.0377807617188, 'zcp_nwot': 214.66846363185357, 'zcp_params': 1514876.0, 'zcp_plain': 0.00639678118750453, 'zcp_snip': 44.79195785522461, 'lat_1080ti_1': 0.5997239113850507, 'lat_1080ti_32': 0.5365121838271364, 'lat_1080ti_64': 0.4863951951941521, 'lat_2080ti_1': 0.6395100078824729, 'lat_2080ti_32': 0.5889615049980718, 'lat_2080ti_64': 0.5316434799770312, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.38982817743384723, 'lat_fpga': 0.40260639824261063, 'lat_gold_6226': 0.23113531988151104, 'lat_gold_6240': 0.3762761645202374, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.3672079932973815, 'lat_raspi4': 0.36380087589086807, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.14173228346456693, 'lat_silver_4114': 0.42067778310855286, 'lat_silver_4210r': 0.41166780667944336, 'lat_titan_rtx_1': 0.5944188349098948, 'lat_titan_rtx_32': 0.547694266171574, 'lat_titan_rtx_64': 0.5789505059664395, 'lat_titanx_1': 0.3085623958664709, 'lat_titanx_32': 0.5258216345467422, 'lat_titanx_64': 0.49576043231320815, 'lat_titanxp_1': 0.5543757022043521, 'lat_titanxp_32': 0.5516097832031751, 'lat_titanxp_64': 0.5022843287589285}
| |
FBNet_1098
|
FBNet
|
1098
|
1098
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_694[FLOAT, 16x3x3x3]
%onnx::Conv_695[FLOAT, 16]
%onnx::Conv_697[FLOAT, 16x16x1x1]
%onnx::Conv_700[FLOAT, 16x1x3x3]
%onnx::Conv_703[FLOAT, 16x16x1x1]
%onnx::Conv_706[FLOAT, 96x16x1x1]
%onnx::Conv_707[FLOAT, 96]
%onnx::Conv_709[FLOAT, 96x1x5x5]
%onnx::Conv_712[FLOAT, 24x96x1x1]
%onnx::Conv_713[FLOAT, 24]
%onnx::Conv_715[FLOAT, 24x12x1x1]
%onnx::Conv_718[FLOAT, 24x1x5x5]
%onnx::Conv_721[FLOAT, 24x12x1x1]
%onnx::Conv_724[FLOAT, 24x12x1x1]
%onnx::Conv_727[FLOAT, 24x1x3x3]
%onnx::Conv_730[FLOAT, 24x12x1x1]
%onnx::Conv_733[FLOAT, 72x24x1x1]
%onnx::Conv_734[FLOAT, 72]
%onnx::Conv_736[FLOAT, 72x1x5x5]
%onnx::Conv_739[FLOAT, 24x72x1x1]
%onnx::Conv_742[FLOAT, 24x12x1x1]
%onnx::Conv_745[FLOAT, 24x1x5x5]
%onnx::Conv_748[FLOAT, 32x12x1x1]
%onnx::Conv_749[FLOAT, 32]
%onnx::Conv_751[FLOAT, 32x16x1x1]
%onnx::Conv_754[FLOAT, 32x1x3x3]
%onnx::Conv_757[FLOAT, 32x16x1x1]
%onnx::Conv_760[FLOAT, 32x16x1x1]
%onnx::Conv_763[FLOAT, 32x1x3x3]
%onnx::Conv_766[FLOAT, 32x16x1x1]
%onnx::Conv_769[FLOAT, 192x32x1x1]
%onnx::Conv_770[FLOAT, 192]
%onnx::Conv_772[FLOAT, 192x1x5x5]
%onnx::Conv_775[FLOAT, 32x192x1x1]
%onnx::Conv_778[FLOAT, 32x32x1x1]
%onnx::Conv_781[FLOAT, 32x1x5x5]
%onnx::Conv_784[FLOAT, 64x32x1x1]
%onnx::Conv_785[FLOAT, 64]
%onnx::Conv_787[FLOAT, 384x64x1x1]
%onnx::Conv_788[FLOAT, 384]
%onnx::Conv_790[FLOAT, 384x1x5x5]
%onnx::Conv_793[FLOAT, 64x384x1x1]
%onnx::Conv_796[FLOAT, 384x64x1x1]
%onnx::Conv_799[FLOAT, 384x1x5x5]
%onnx::Conv_802[FLOAT, 64x384x1x1]
%onnx::Conv_805[FLOAT, 192x64x1x1]
%onnx::Conv_808[FLOAT, 192x1x5x5]
%onnx::Conv_811[FLOAT, 112x192x1x1]
%onnx::Conv_812[FLOAT, 112]
%onnx::Conv_814[FLOAT, 112x112x1x1]
%onnx::Conv_817[FLOAT, 112x1x3x3]
%onnx::Conv_820[FLOAT, 112x112x1x1]
%onnx::Conv_823[FLOAT, 112x56x1x1]
%onnx::Conv_826[FLOAT, 112x1x5x5]
%onnx::Conv_829[FLOAT, 112x56x1x1]
%onnx::Conv_832[FLOAT, 112x56x1x1]
%onnx::Conv_835[FLOAT, 112x1x3x3]
%onnx::Conv_838[FLOAT, 184x56x1x1]
%onnx::Conv_839[FLOAT, 184]
%onnx::Conv_841[FLOAT, 1104x184x1x1]
%onnx::Conv_842[FLOAT, 1104]
%onnx::Conv_844[FLOAT, 1104x1x5x5]
%onnx::Conv_847[FLOAT, 184x1104x1x1]
%onnx::Conv_850[FLOAT, 184x184x1x1]
%onnx::Conv_853[FLOAT, 184x1x3x3]
%onnx::Conv_856[FLOAT, 184x184x1x1]
%onnx::Conv_859[FLOAT, 552x184x1x1]
%onnx::Conv_860[FLOAT, 552]
%onnx::Conv_862[FLOAT, 552x1x3x3]
%onnx::Conv_865[FLOAT, 184x552x1x1]
%onnx::Conv_868[FLOAT, 1104x184x1x1]
%onnx::Conv_871[FLOAT, 1104x1x5x5]
%onnx::Conv_874[FLOAT, 352x1104x1x1]
%onnx::Conv_875[FLOAT, 352]
%onnx::Conv_877[FLOAT, 1504x352x1x1]
%onnx::Conv_878[FLOAT, 1504]
) {
%onnx::Conv_872 = Identity(%onnx::Conv_842)
%onnx::Conv_869 = Identity(%onnx::Conv_842)
%onnx::Conv_866 = Identity(%onnx::Conv_839)
%onnx::Conv_863 = Identity(%onnx::Conv_860)
%onnx::Conv_857 = Identity(%onnx::Conv_839)
%onnx::Conv_854 = Identity(%onnx::Conv_839)
%onnx::Conv_851 = Identity(%onnx::Conv_839)
%onnx::Conv_848 = Identity(%onnx::Conv_839)
%onnx::Conv_845 = Identity(%onnx::Conv_842)
%onnx::Conv_836 = Identity(%onnx::Conv_812)
%onnx::Conv_833 = Identity(%onnx::Conv_812)
%onnx::Conv_830 = Identity(%onnx::Conv_812)
%onnx::Conv_827 = Identity(%onnx::Conv_812)
%onnx::Conv_824 = Identity(%onnx::Conv_812)
%onnx::Conv_821 = Identity(%onnx::Conv_812)
%onnx::Conv_818 = Identity(%onnx::Conv_812)
%onnx::Conv_815 = Identity(%onnx::Conv_812)
%onnx::Conv_809 = Identity(%onnx::Conv_770)
%onnx::Conv_806 = Identity(%onnx::Conv_770)
%onnx::Conv_803 = Identity(%onnx::Conv_785)
%onnx::Conv_800 = Identity(%onnx::Conv_788)
%onnx::Conv_797 = Identity(%onnx::Conv_788)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_788)
%onnx::Conv_782 = Identity(%onnx::Conv_749)
%onnx::Conv_779 = Identity(%onnx::Conv_749)
%onnx::Conv_776 = Identity(%onnx::Conv_749)
%onnx::Conv_773 = Identity(%onnx::Conv_770)
%onnx::Conv_767 = Identity(%onnx::Conv_749)
%onnx::Conv_764 = Identity(%onnx::Conv_749)
%onnx::Conv_761 = Identity(%onnx::Conv_749)
%onnx::Conv_758 = Identity(%onnx::Conv_749)
%onnx::Conv_755 = Identity(%onnx::Conv_749)
%onnx::Conv_752 = Identity(%onnx::Conv_749)
%onnx::Conv_746 = Identity(%onnx::Conv_713)
%onnx::Conv_743 = Identity(%onnx::Conv_713)
%onnx::Conv_740 = Identity(%onnx::Conv_713)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_731 = Identity(%onnx::Conv_713)
%onnx::Conv_728 = Identity(%onnx::Conv_713)
%onnx::Conv_725 = Identity(%onnx::Conv_713)
%onnx::Conv_722 = Identity(%onnx::Conv_713)
%onnx::Conv_719 = Identity(%onnx::Conv_713)
%onnx::Conv_716 = Identity(%onnx::Conv_713)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_695)
%onnx::Conv_701 = Identity(%onnx::Conv_695)
%onnx::Conv_698 = Identity(%onnx::Conv_695)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_694, %onnx::Conv_695)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_862, %onnx::Conv_863)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_865, %onnx::Conv_866)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_868, %onnx::Conv_869)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_871, %onnx::Conv_872)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_874, %onnx::Conv_875)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_877, %onnx::Conv_878)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%692 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %692
}
|
val_accuracy
| 0 | 66,857,856 | 2,291,012 |
{'zcp_synflow': 73.42402287871589, 'zcp_zen': 65.59107208251953, 'zcp_epe_nas': 22.477010102614507, 'zcp_fisher': 0.09904425591230392, 'zcp_flops': 66857856.0, 'zcp_grad_norm': 25.29189109802246, 'zcp_grasp': -0.15439224243164062, 'zcp_jacov': -16.063505541649015, 'zcp_l2_norm': 607.7509765625, 'zcp_nwot': 210.62175135780802, 'zcp_params': 2291012.0, 'zcp_plain': 0.0047743795439600945, 'zcp_snip': 39.9777946472168, 'lat_1080ti_1': 0.6400286114707401, 'lat_1080ti_32': 0.5688391427869921, 'lat_1080ti_64': 0.4768823295547228, 'lat_2080ti_1': 0.6461288714383789, 'lat_2080ti_32': 0.5812553932243636, 'lat_2080ti_64': 0.4593561174608493, 'lat_essential_ph_1': 0.5849056603773585, 'lat_eyeriss': 0.4965250398841068, 'lat_fpga': 0.4567233002503487, 'lat_gold_6226': 0.4180017163137092, 'lat_gold_6240': 0.7956583701566844, 'lat_pixel2': 0.32608695652173914, 'lat_pixel3': 0.5240825875922852, 'lat_raspi4': 0.5612802343247039, 'lat_samsung_a50': 0.2, 'lat_samsung_s7': 0.18110236220472442, 'lat_silver_4114': 0.6431817220979972, 'lat_silver_4210r': 0.6708811506245295, 'lat_titan_rtx_1': 0.6222383566318345, 'lat_titan_rtx_32': 0.5902855811162704, 'lat_titan_rtx_64': 0.5007407763415807, 'lat_titanx_1': 0.32783557760197896, 'lat_titanx_32': 0.5262421446440056, 'lat_titanx_64': 0.47220034949706435, 'lat_titanxp_1': 0.5836022974305556, 'lat_titanxp_32': 0.572821264362235, 'lat_titanxp_64': 0.4689836422244701}
| |
FBNet_2843
|
FBNet
|
2843
|
2843
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_589[FLOAT, 16x3x3x3]
%onnx::Conv_590[FLOAT, 16]
%onnx::Conv_592[FLOAT, 96x16x1x1]
%onnx::Conv_593[FLOAT, 96]
%onnx::Conv_595[FLOAT, 96x1x5x5]
%onnx::Conv_598[FLOAT, 16x96x1x1]
%onnx::Conv_601[FLOAT, 16x8x1x1]
%onnx::Conv_604[FLOAT, 16x1x5x5]
%onnx::Conv_607[FLOAT, 24x8x1x1]
%onnx::Conv_608[FLOAT, 24]
%onnx::Conv_610[FLOAT, 72x24x1x1]
%onnx::Conv_611[FLOAT, 72]
%onnx::Conv_613[FLOAT, 72x1x3x3]
%onnx::Conv_616[FLOAT, 24x72x1x1]
%onnx::Conv_619[FLOAT, 72x24x1x1]
%onnx::Conv_622[FLOAT, 72x1x3x3]
%onnx::Conv_625[FLOAT, 32x72x1x1]
%onnx::Conv_626[FLOAT, 32]
%onnx::Conv_628[FLOAT, 96x32x1x1]
%onnx::Conv_631[FLOAT, 96x1x3x3]
%onnx::Conv_634[FLOAT, 32x96x1x1]
%onnx::Conv_637[FLOAT, 32x16x1x1]
%onnx::Conv_640[FLOAT, 32x1x3x3]
%onnx::Conv_643[FLOAT, 32x16x1x1]
%onnx::Conv_646[FLOAT, 32x16x1x1]
%onnx::Conv_649[FLOAT, 32x1x3x3]
%onnx::Conv_652[FLOAT, 64x16x1x1]
%onnx::Conv_653[FLOAT, 64]
%onnx::Conv_655[FLOAT, 64x32x1x1]
%onnx::Conv_658[FLOAT, 64x1x3x3]
%onnx::Conv_661[FLOAT, 64x32x1x1]
%onnx::Conv_664[FLOAT, 192x64x1x1]
%onnx::Conv_665[FLOAT, 192]
%onnx::Conv_667[FLOAT, 192x1x5x5]
%onnx::Conv_670[FLOAT, 64x192x1x1]
%onnx::Conv_673[FLOAT, 64x64x1x1]
%onnx::Conv_676[FLOAT, 64x1x5x5]
%onnx::Conv_679[FLOAT, 112x64x1x1]
%onnx::Conv_680[FLOAT, 112]
%onnx::Conv_682[FLOAT, 112x56x1x1]
%onnx::Conv_685[FLOAT, 112x1x5x5]
%onnx::Conv_688[FLOAT, 112x56x1x1]
%onnx::Conv_691[FLOAT, 112x56x1x1]
%onnx::Conv_694[FLOAT, 112x1x5x5]
%onnx::Conv_697[FLOAT, 112x56x1x1]
%onnx::Conv_700[FLOAT, 112x56x1x1]
%onnx::Conv_703[FLOAT, 112x1x5x5]
%onnx::Conv_706[FLOAT, 112x56x1x1]
%onnx::Conv_709[FLOAT, 184x112x1x1]
%onnx::Conv_710[FLOAT, 184]
%onnx::Conv_712[FLOAT, 184x92x1x1]
%onnx::Conv_715[FLOAT, 184x1x5x5]
%onnx::Conv_718[FLOAT, 184x92x1x1]
%onnx::Conv_721[FLOAT, 1104x184x1x1]
%onnx::Conv_722[FLOAT, 1104]
%onnx::Conv_724[FLOAT, 1104x1x5x5]
%onnx::Conv_727[FLOAT, 352x1104x1x1]
%onnx::Conv_728[FLOAT, 352]
%onnx::Conv_730[FLOAT, 1504x352x1x1]
%onnx::Conv_731[FLOAT, 1504]
) {
%onnx::Conv_725 = Identity(%onnx::Conv_722)
%onnx::Conv_719 = Identity(%onnx::Conv_710)
%onnx::Conv_716 = Identity(%onnx::Conv_710)
%onnx::Conv_713 = Identity(%onnx::Conv_710)
%onnx::Conv_707 = Identity(%onnx::Conv_680)
%onnx::Conv_704 = Identity(%onnx::Conv_680)
%onnx::Conv_701 = Identity(%onnx::Conv_680)
%onnx::Conv_698 = Identity(%onnx::Conv_680)
%onnx::Conv_695 = Identity(%onnx::Conv_680)
%onnx::Conv_692 = Identity(%onnx::Conv_680)
%onnx::Conv_689 = Identity(%onnx::Conv_680)
%onnx::Conv_686 = Identity(%onnx::Conv_680)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_677 = Identity(%onnx::Conv_653)
%onnx::Conv_674 = Identity(%onnx::Conv_653)
%onnx::Conv_671 = Identity(%onnx::Conv_653)
%onnx::Conv_668 = Identity(%onnx::Conv_665)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_650 = Identity(%onnx::Conv_626)
%onnx::Conv_647 = Identity(%onnx::Conv_626)
%onnx::Conv_644 = Identity(%onnx::Conv_626)
%onnx::Conv_641 = Identity(%onnx::Conv_626)
%onnx::Conv_638 = Identity(%onnx::Conv_626)
%onnx::Conv_635 = Identity(%onnx::Conv_626)
%onnx::Conv_632 = Identity(%onnx::Conv_593)
%onnx::Conv_629 = Identity(%onnx::Conv_593)
%onnx::Conv_623 = Identity(%onnx::Conv_611)
%onnx::Conv_620 = Identity(%onnx::Conv_611)
%onnx::Conv_617 = Identity(%onnx::Conv_608)
%onnx::Conv_614 = Identity(%onnx::Conv_611)
%onnx::Conv_605 = Identity(%onnx::Conv_590)
%onnx::Conv_602 = Identity(%onnx::Conv_590)
%onnx::Conv_599 = Identity(%onnx::Conv_590)
%onnx::Conv_596 = Identity(%onnx::Conv_593)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_589, %onnx::Conv_590)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_592, %onnx::Conv_593)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_595, %onnx::Conv_596)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_598, %onnx::Conv_599)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_601, %onnx::Conv_602)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_604, %onnx::Conv_605)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_607, %onnx::Conv_608)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_610, %onnx::Conv_611)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_613, %onnx::Conv_614)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%587 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %587
}
|
val_accuracy
| 0 | 43,357,184 | 1,492,332 |
{'zcp_synflow': 55.57232321087233, 'zcp_zen': 48.285186767578125, 'zcp_epe_nas': 16.267457471377362, 'zcp_fisher': 0.03447488695383072, 'zcp_flops': 43357184.0, 'zcp_grad_norm': 14.918638229370117, 'zcp_grasp': -0.02040386199951172, 'zcp_jacov': -16.04887313930687, 'zcp_l2_norm': 401.61785888671875, 'zcp_nwot': 206.13649271280485, 'zcp_params': 1492332.0, 'zcp_plain': 0.00800784770399332, 'zcp_snip': 21.300729751586914, 'lat_1080ti_1': 0.17702953719559125, 'lat_1080ti_32': 0.21085629729560146, 'lat_1080ti_64': 0.19080893537338867, 'lat_2080ti_1': 0.21823157172605584, 'lat_2080ti_32': 0.2231529300429923, 'lat_2080ti_64': 0.2020433027750769, 'lat_essential_ph_1': 0.05660377358490566, 'lat_eyeriss': 0.14708525113234175, 'lat_fpga': 0.146102771648245, 'lat_gold_6226': 0.11500341405260128, 'lat_gold_6240': 0.0947100574090617, 'lat_pixel2': 0.13043478260869565, 'lat_pixel3': 0.193049588591861, 'lat_raspi4': 0.2630997746357416, 'lat_samsung_a50': 0.05263157894736842, 'lat_samsung_s7': 0.047244094488188976, 'lat_silver_4114': 0.0749132333314436, 'lat_silver_4210r': 0.0, 'lat_titan_rtx_1': 0.20022188994064774, 'lat_titan_rtx_32': 0.21367117099327257, 'lat_titan_rtx_64': 0.1757947651546002, 'lat_titanx_1': 0.0983984420623223, 'lat_titanx_32': 0.17668092354769332, 'lat_titanx_64': 0.16879467099159612, 'lat_titanxp_1': 0.18091803768905682, 'lat_titanxp_32': 0.1697593532586842, 'lat_titanxp_64': 0.17944241384103768}
| |
FBNet_2336
|
FBNet
|
2336
|
2336
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_586[FLOAT, 16x3x3x3]
%onnx::Conv_587[FLOAT, 16]
%onnx::Conv_589[FLOAT, 48x16x1x1]
%onnx::Conv_590[FLOAT, 48]
%onnx::Conv_592[FLOAT, 48x1x5x5]
%onnx::Conv_595[FLOAT, 16x48x1x1]
%onnx::Conv_598[FLOAT, 48x16x1x1]
%onnx::Conv_601[FLOAT, 48x1x3x3]
%onnx::Conv_604[FLOAT, 24x48x1x1]
%onnx::Conv_605[FLOAT, 24]
%onnx::Conv_607[FLOAT, 24x12x1x1]
%onnx::Conv_610[FLOAT, 24x1x5x5]
%onnx::Conv_613[FLOAT, 32x12x1x1]
%onnx::Conv_614[FLOAT, 32]
%onnx::Conv_616[FLOAT, 32x16x1x1]
%onnx::Conv_619[FLOAT, 32x1x5x5]
%onnx::Conv_622[FLOAT, 32x16x1x1]
%onnx::Conv_625[FLOAT, 96x32x1x1]
%onnx::Conv_626[FLOAT, 96]
%onnx::Conv_628[FLOAT, 96x1x5x5]
%onnx::Conv_631[FLOAT, 32x96x1x1]
%onnx::Conv_634[FLOAT, 96x32x1x1]
%onnx::Conv_637[FLOAT, 96x1x5x5]
%onnx::Conv_640[FLOAT, 64x96x1x1]
%onnx::Conv_641[FLOAT, 64]
%onnx::Conv_643[FLOAT, 192x64x1x1]
%onnx::Conv_644[FLOAT, 192]
%onnx::Conv_646[FLOAT, 192x1x5x5]
%onnx::Conv_649[FLOAT, 64x192x1x1]
%onnx::Conv_652[FLOAT, 192x64x1x1]
%onnx::Conv_655[FLOAT, 192x1x3x3]
%onnx::Conv_658[FLOAT, 64x192x1x1]
%onnx::Conv_661[FLOAT, 192x64x1x1]
%onnx::Conv_664[FLOAT, 192x1x3x3]
%onnx::Conv_667[FLOAT, 64x192x1x1]
%onnx::Conv_670[FLOAT, 384x64x1x1]
%onnx::Conv_671[FLOAT, 384]
%onnx::Conv_673[FLOAT, 384x1x5x5]
%onnx::Conv_676[FLOAT, 112x384x1x1]
%onnx::Conv_677[FLOAT, 112]
%onnx::Conv_679[FLOAT, 672x112x1x1]
%onnx::Conv_680[FLOAT, 672]
%onnx::Conv_682[FLOAT, 672x1x3x3]
%onnx::Conv_685[FLOAT, 112x672x1x1]
%onnx::Conv_688[FLOAT, 112x56x1x1]
%onnx::Conv_691[FLOAT, 112x1x5x5]
%onnx::Conv_694[FLOAT, 112x56x1x1]
%onnx::Conv_697[FLOAT, 112x56x1x1]
%onnx::Conv_700[FLOAT, 112x1x5x5]
%onnx::Conv_703[FLOAT, 112x56x1x1]
%onnx::Conv_706[FLOAT, 112x112x1x1]
%onnx::Conv_709[FLOAT, 112x1x5x5]
%onnx::Conv_712[FLOAT, 184x112x1x1]
%onnx::Conv_713[FLOAT, 184]
%onnx::Conv_715[FLOAT, 184x184x1x1]
%onnx::Conv_718[FLOAT, 184x1x5x5]
%onnx::Conv_721[FLOAT, 184x184x1x1]
%onnx::Conv_724[FLOAT, 1104x184x1x1]
%onnx::Conv_725[FLOAT, 1104]
%onnx::Conv_727[FLOAT, 1104x1x5x5]
%onnx::Conv_730[FLOAT, 184x1104x1x1]
%onnx::Conv_733[FLOAT, 1104x184x1x1]
%onnx::Conv_736[FLOAT, 1104x1x5x5]
%onnx::Conv_739[FLOAT, 184x1104x1x1]
%onnx::Conv_742[FLOAT, 184x184x1x1]
%onnx::Conv_745[FLOAT, 184x1x5x5]
%onnx::Conv_748[FLOAT, 352x184x1x1]
%onnx::Conv_749[FLOAT, 352]
%onnx::Conv_751[FLOAT, 1504x352x1x1]
%onnx::Conv_752[FLOAT, 1504]
) {
%onnx::Conv_746 = Identity(%onnx::Conv_713)
%onnx::Conv_743 = Identity(%onnx::Conv_713)
%onnx::Conv_740 = Identity(%onnx::Conv_713)
%onnx::Conv_737 = Identity(%onnx::Conv_725)
%onnx::Conv_734 = Identity(%onnx::Conv_725)
%onnx::Conv_731 = Identity(%onnx::Conv_713)
%onnx::Conv_728 = Identity(%onnx::Conv_725)
%onnx::Conv_722 = Identity(%onnx::Conv_713)
%onnx::Conv_719 = Identity(%onnx::Conv_713)
%onnx::Conv_716 = Identity(%onnx::Conv_713)
%onnx::Conv_710 = Identity(%onnx::Conv_677)
%onnx::Conv_707 = Identity(%onnx::Conv_677)
%onnx::Conv_704 = Identity(%onnx::Conv_677)
%onnx::Conv_701 = Identity(%onnx::Conv_677)
%onnx::Conv_698 = Identity(%onnx::Conv_677)
%onnx::Conv_695 = Identity(%onnx::Conv_677)
%onnx::Conv_692 = Identity(%onnx::Conv_677)
%onnx::Conv_689 = Identity(%onnx::Conv_677)
%onnx::Conv_686 = Identity(%onnx::Conv_677)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_674 = Identity(%onnx::Conv_671)
%onnx::Conv_668 = Identity(%onnx::Conv_641)
%onnx::Conv_665 = Identity(%onnx::Conv_644)
%onnx::Conv_662 = Identity(%onnx::Conv_644)
%onnx::Conv_659 = Identity(%onnx::Conv_641)
%onnx::Conv_656 = Identity(%onnx::Conv_644)
%onnx::Conv_653 = Identity(%onnx::Conv_644)
%onnx::Conv_650 = Identity(%onnx::Conv_641)
%onnx::Conv_647 = Identity(%onnx::Conv_644)
%onnx::Conv_638 = Identity(%onnx::Conv_626)
%onnx::Conv_635 = Identity(%onnx::Conv_626)
%onnx::Conv_632 = Identity(%onnx::Conv_614)
%onnx::Conv_629 = Identity(%onnx::Conv_626)
%onnx::Conv_623 = Identity(%onnx::Conv_614)
%onnx::Conv_620 = Identity(%onnx::Conv_614)
%onnx::Conv_617 = Identity(%onnx::Conv_614)
%onnx::Conv_611 = Identity(%onnx::Conv_605)
%onnx::Conv_608 = Identity(%onnx::Conv_605)
%onnx::Conv_602 = Identity(%onnx::Conv_590)
%onnx::Conv_599 = Identity(%onnx::Conv_590)
%onnx::Conv_596 = Identity(%onnx::Conv_587)
%onnx::Conv_593 = Identity(%onnx::Conv_590)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_586, %onnx::Conv_587)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_589, %onnx::Conv_590)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_592, %onnx::Conv_593)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_595, %onnx::Conv_596)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_598, %onnx::Conv_599)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_601, %onnx::Conv_602)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_604, %onnx::Conv_605)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_607, %onnx::Conv_608)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_610, %onnx::Conv_611)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_613, %onnx::Conv_614)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%584 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %584
}
|
val_accuracy
| 0 | 60,860,800 | 2,160,796 |
{'zcp_synflow': 72.01676178860092, 'zcp_zen': 65.69124603271484, 'zcp_epe_nas': 22.483627315153985, 'zcp_fisher': 0.06478246301412582, 'zcp_flops': 60860800.0, 'zcp_grad_norm': 19.239295959472656, 'zcp_grasp': 0.0014181137084960938, 'zcp_jacov': -16.055551841852747, 'zcp_l2_norm': 614.4989624023438, 'zcp_nwot': 205.42873887740296, 'zcp_params': 2160796.0, 'zcp_plain': 0.005228728987276554, 'zcp_snip': 36.536582946777344, 'lat_1080ti_1': 0.34663015366544236, 'lat_1080ti_32': 0.24863644674171598, 'lat_1080ti_64': 0.17215457777305226, 'lat_2080ti_1': 0.40349540776533327, 'lat_2080ti_32': 0.2742770444559988, 'lat_2080ti_64': 0.17232756892697085, 'lat_essential_ph_1': 0.39622641509433965, 'lat_eyeriss': 0.35066840261859805, 'lat_fpga': 0.40316052998743307, 'lat_gold_6226': 0.40643742064079985, 'lat_gold_6240': 0.46565170823567686, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.319058946558628, 'lat_raspi4': 0.34966316964695554, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.15748031496062992, 'lat_silver_4114': 0.4338351522589657, 'lat_silver_4210r': 0.42963588705728856, 'lat_titan_rtx_1': 0.46324176684494894, 'lat_titan_rtx_32': 0.27605951610381624, 'lat_titan_rtx_64': 0.18155050530583938, 'lat_titanx_1': 0.1953857232404574, 'lat_titanx_32': 0.2420302705819525, 'lat_titanx_64': 0.16255832587316854, 'lat_titanxp_1': 0.35579165523610035, 'lat_titanxp_32': 0.2417137394874127, 'lat_titanxp_64': 0.1856070152055346}
| |
FBNet_243
|
FBNet
|
243
|
243
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_615[FLOAT, 16x3x3x3]
%onnx::Conv_616[FLOAT, 16]
%onnx::Conv_618[FLOAT, 96x16x1x1]
%onnx::Conv_619[FLOAT, 96]
%onnx::Conv_621[FLOAT, 96x1x5x5]
%onnx::Conv_624[FLOAT, 16x96x1x1]
%onnx::Conv_627[FLOAT, 48x16x1x1]
%onnx::Conv_628[FLOAT, 48]
%onnx::Conv_630[FLOAT, 48x1x5x5]
%onnx::Conv_633[FLOAT, 24x48x1x1]
%onnx::Conv_634[FLOAT, 24]
%onnx::Conv_636[FLOAT, 144x24x1x1]
%onnx::Conv_637[FLOAT, 144]
%onnx::Conv_639[FLOAT, 144x1x5x5]
%onnx::Conv_642[FLOAT, 24x144x1x1]
%onnx::Conv_645[FLOAT, 72x24x1x1]
%onnx::Conv_646[FLOAT, 72]
%onnx::Conv_648[FLOAT, 72x1x5x5]
%onnx::Conv_651[FLOAT, 24x72x1x1]
%onnx::Conv_654[FLOAT, 24x24x1x1]
%onnx::Conv_657[FLOAT, 24x1x5x5]
%onnx::Conv_660[FLOAT, 24x24x1x1]
%onnx::Conv_663[FLOAT, 24x12x1x1]
%onnx::Conv_666[FLOAT, 24x1x3x3]
%onnx::Conv_669[FLOAT, 32x12x1x1]
%onnx::Conv_670[FLOAT, 32]
%onnx::Conv_672[FLOAT, 32x16x1x1]
%onnx::Conv_675[FLOAT, 32x1x5x5]
%onnx::Conv_678[FLOAT, 32x16x1x1]
%onnx::Conv_681[FLOAT, 96x32x1x1]
%onnx::Conv_684[FLOAT, 96x1x5x5]
%onnx::Conv_687[FLOAT, 32x96x1x1]
%onnx::Conv_690[FLOAT, 32x32x1x1]
%onnx::Conv_693[FLOAT, 32x1x5x5]
%onnx::Conv_696[FLOAT, 32x32x1x1]
%onnx::Conv_699[FLOAT, 64x32x1x1]
%onnx::Conv_700[FLOAT, 64]
%onnx::Conv_702[FLOAT, 192x64x1x1]
%onnx::Conv_703[FLOAT, 192]
%onnx::Conv_705[FLOAT, 192x1x5x5]
%onnx::Conv_708[FLOAT, 64x192x1x1]
%onnx::Conv_711[FLOAT, 64x64x1x1]
%onnx::Conv_714[FLOAT, 64x1x5x5]
%onnx::Conv_717[FLOAT, 64x64x1x1]
%onnx::Conv_720[FLOAT, 64x64x1x1]
%onnx::Conv_723[FLOAT, 64x1x3x3]
%onnx::Conv_726[FLOAT, 64x64x1x1]
%onnx::Conv_729[FLOAT, 192x64x1x1]
%onnx::Conv_732[FLOAT, 192x1x3x3]
%onnx::Conv_735[FLOAT, 112x192x1x1]
%onnx::Conv_736[FLOAT, 112]
%onnx::Conv_738[FLOAT, 672x112x1x1]
%onnx::Conv_739[FLOAT, 672]
%onnx::Conv_741[FLOAT, 672x1x3x3]
%onnx::Conv_744[FLOAT, 112x672x1x1]
%onnx::Conv_747[FLOAT, 672x112x1x1]
%onnx::Conv_750[FLOAT, 672x1x3x3]
%onnx::Conv_753[FLOAT, 112x672x1x1]
%onnx::Conv_756[FLOAT, 184x112x1x1]
%onnx::Conv_757[FLOAT, 184]
%onnx::Conv_759[FLOAT, 1104x184x1x1]
%onnx::Conv_760[FLOAT, 1104]
%onnx::Conv_762[FLOAT, 1104x1x5x5]
%onnx::Conv_765[FLOAT, 184x1104x1x1]
%onnx::Conv_768[FLOAT, 1104x184x1x1]
%onnx::Conv_771[FLOAT, 1104x1x5x5]
%onnx::Conv_774[FLOAT, 184x1104x1x1]
%onnx::Conv_777[FLOAT, 1104x184x1x1]
%onnx::Conv_780[FLOAT, 1104x1x3x3]
%onnx::Conv_783[FLOAT, 184x1104x1x1]
%onnx::Conv_786[FLOAT, 184x92x1x1]
%onnx::Conv_789[FLOAT, 184x1x3x3]
%onnx::Conv_792[FLOAT, 352x92x1x1]
%onnx::Conv_793[FLOAT, 352]
%onnx::Conv_795[FLOAT, 1504x352x1x1]
%onnx::Conv_796[FLOAT, 1504]
) {
%onnx::Conv_790 = Identity(%onnx::Conv_757)
%onnx::Conv_787 = Identity(%onnx::Conv_757)
%onnx::Conv_784 = Identity(%onnx::Conv_757)
%onnx::Conv_781 = Identity(%onnx::Conv_760)
%onnx::Conv_778 = Identity(%onnx::Conv_760)
%onnx::Conv_775 = Identity(%onnx::Conv_757)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_757)
%onnx::Conv_763 = Identity(%onnx::Conv_760)
%onnx::Conv_754 = Identity(%onnx::Conv_736)
%onnx::Conv_751 = Identity(%onnx::Conv_739)
%onnx::Conv_748 = Identity(%onnx::Conv_739)
%onnx::Conv_745 = Identity(%onnx::Conv_736)
%onnx::Conv_742 = Identity(%onnx::Conv_739)
%onnx::Conv_733 = Identity(%onnx::Conv_703)
%onnx::Conv_730 = Identity(%onnx::Conv_703)
%onnx::Conv_727 = Identity(%onnx::Conv_700)
%onnx::Conv_724 = Identity(%onnx::Conv_700)
%onnx::Conv_721 = Identity(%onnx::Conv_700)
%onnx::Conv_718 = Identity(%onnx::Conv_700)
%onnx::Conv_715 = Identity(%onnx::Conv_700)
%onnx::Conv_712 = Identity(%onnx::Conv_700)
%onnx::Conv_709 = Identity(%onnx::Conv_700)
%onnx::Conv_706 = Identity(%onnx::Conv_703)
%onnx::Conv_697 = Identity(%onnx::Conv_670)
%onnx::Conv_694 = Identity(%onnx::Conv_670)
%onnx::Conv_691 = Identity(%onnx::Conv_670)
%onnx::Conv_688 = Identity(%onnx::Conv_670)
%onnx::Conv_685 = Identity(%onnx::Conv_619)
%onnx::Conv_682 = Identity(%onnx::Conv_619)
%onnx::Conv_679 = Identity(%onnx::Conv_670)
%onnx::Conv_676 = Identity(%onnx::Conv_670)
%onnx::Conv_673 = Identity(%onnx::Conv_670)
%onnx::Conv_667 = Identity(%onnx::Conv_634)
%onnx::Conv_664 = Identity(%onnx::Conv_634)
%onnx::Conv_661 = Identity(%onnx::Conv_634)
%onnx::Conv_658 = Identity(%onnx::Conv_634)
%onnx::Conv_655 = Identity(%onnx::Conv_634)
%onnx::Conv_652 = Identity(%onnx::Conv_634)
%onnx::Conv_649 = Identity(%onnx::Conv_646)
%onnx::Conv_643 = Identity(%onnx::Conv_634)
%onnx::Conv_640 = Identity(%onnx::Conv_637)
%onnx::Conv_631 = Identity(%onnx::Conv_628)
%onnx::Conv_625 = Identity(%onnx::Conv_616)
%onnx::Conv_622 = Identity(%onnx::Conv_619)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_615, %onnx::Conv_616)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.15/Add_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%613 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %613
}
|
val_accuracy
| 0 | 90,506,240 | 2,505,300 |
{'zcp_synflow': 79.68283279572847, 'zcp_zen': 68.6905746459961, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.16959168016910553, 'zcp_flops': 90506240.0, 'zcp_grad_norm': 27.98672103881836, 'zcp_grasp': 0.0099334716796875, 'zcp_jacov': -16.065995309725096, 'zcp_l2_norm': 667.7710571289062, 'zcp_nwot': 215.62950465072447, 'zcp_params': 2505300.0, 'zcp_plain': -0.00433104345574975, 'zcp_snip': 50.09499740600586, 'lat_1080ti_1': 0.5370052005015579, 'lat_1080ti_32': 0.5983621433103711, 'lat_1080ti_64': 0.6708604560973488, 'lat_2080ti_1': 0.5397725369164416, 'lat_2080ti_32': 0.5696684165449023, 'lat_2080ti_64': 0.6067707625225581, 'lat_essential_ph_1': 0.4716981132075472, 'lat_eyeriss': 0.7329048466066417, 'lat_fpga': 0.7665720024935927, 'lat_gold_6226': 0.5619549441191015, 'lat_gold_6240': 0.7255765643713619, 'lat_pixel2': 0.5434782608695652, 'lat_pixel3': 0.7553170649867303, 'lat_raspi4': 0.807455512632611, 'lat_samsung_a50': 0.30526315789473685, 'lat_samsung_s7': 0.7007874015748031, 'lat_silver_4114': 0.7889958355774221, 'lat_silver_4210r': 0.730771285505843, 'lat_titan_rtx_1': 0.5232612105040814, 'lat_titan_rtx_32': 0.5526301645343166, 'lat_titan_rtx_64': 0.6093669499461499, 'lat_titanx_1': 0.28982922436164843, 'lat_titanx_32': 0.618224950107599, 'lat_titanx_64': 0.7186490464697766, 'lat_titanxp_1': 0.49478298307854074, 'lat_titanxp_32': 0.6310221153186925, 'lat_titanxp_64': 0.6562218914659486}
| |
FBNet_2805
|
FBNet
|
2805
|
2805
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_613[FLOAT, 16x3x3x3]
%onnx::Conv_614[FLOAT, 16]
%onnx::Conv_616[FLOAT, 16x16x1x1]
%onnx::Conv_619[FLOAT, 16x1x5x5]
%onnx::Conv_622[FLOAT, 16x16x1x1]
%onnx::Conv_625[FLOAT, 96x16x1x1]
%onnx::Conv_626[FLOAT, 96]
%onnx::Conv_628[FLOAT, 96x1x5x5]
%onnx::Conv_631[FLOAT, 24x96x1x1]
%onnx::Conv_632[FLOAT, 24]
%onnx::Conv_634[FLOAT, 24x24x1x1]
%onnx::Conv_637[FLOAT, 24x1x3x3]
%onnx::Conv_640[FLOAT, 24x24x1x1]
%onnx::Conv_643[FLOAT, 72x24x1x1]
%onnx::Conv_644[FLOAT, 72]
%onnx::Conv_646[FLOAT, 72x1x5x5]
%onnx::Conv_649[FLOAT, 24x72x1x1]
%onnx::Conv_652[FLOAT, 24x24x1x1]
%onnx::Conv_655[FLOAT, 24x1x5x5]
%onnx::Conv_658[FLOAT, 24x24x1x1]
%onnx::Conv_661[FLOAT, 24x24x1x1]
%onnx::Conv_664[FLOAT, 24x1x5x5]
%onnx::Conv_667[FLOAT, 32x24x1x1]
%onnx::Conv_668[FLOAT, 32]
%onnx::Conv_670[FLOAT, 96x32x1x1]
%onnx::Conv_673[FLOAT, 96x1x5x5]
%onnx::Conv_676[FLOAT, 32x96x1x1]
%onnx::Conv_679[FLOAT, 192x32x1x1]
%onnx::Conv_680[FLOAT, 192]
%onnx::Conv_682[FLOAT, 192x1x3x3]
%onnx::Conv_685[FLOAT, 32x192x1x1]
%onnx::Conv_688[FLOAT, 32x16x1x1]
%onnx::Conv_691[FLOAT, 32x1x5x5]
%onnx::Conv_694[FLOAT, 32x16x1x1]
%onnx::Conv_697[FLOAT, 64x32x1x1]
%onnx::Conv_698[FLOAT, 64]
%onnx::Conv_700[FLOAT, 384x64x1x1]
%onnx::Conv_701[FLOAT, 384]
%onnx::Conv_703[FLOAT, 384x1x5x5]
%onnx::Conv_706[FLOAT, 64x384x1x1]
%onnx::Conv_709[FLOAT, 64x64x1x1]
%onnx::Conv_712[FLOAT, 64x1x3x3]
%onnx::Conv_715[FLOAT, 64x64x1x1]
%onnx::Conv_718[FLOAT, 192x64x1x1]
%onnx::Conv_721[FLOAT, 192x1x5x5]
%onnx::Conv_724[FLOAT, 64x192x1x1]
%onnx::Conv_727[FLOAT, 64x64x1x1]
%onnx::Conv_730[FLOAT, 64x1x3x3]
%onnx::Conv_733[FLOAT, 112x64x1x1]
%onnx::Conv_734[FLOAT, 112]
%onnx::Conv_736[FLOAT, 112x112x1x1]
%onnx::Conv_739[FLOAT, 112x1x5x5]
%onnx::Conv_742[FLOAT, 112x112x1x1]
%onnx::Conv_745[FLOAT, 336x112x1x1]
%onnx::Conv_746[FLOAT, 336]
%onnx::Conv_748[FLOAT, 336x1x5x5]
%onnx::Conv_751[FLOAT, 112x336x1x1]
%onnx::Conv_754[FLOAT, 112x112x1x1]
%onnx::Conv_757[FLOAT, 112x1x5x5]
%onnx::Conv_760[FLOAT, 112x112x1x1]
%onnx::Conv_763[FLOAT, 336x112x1x1]
%onnx::Conv_766[FLOAT, 336x1x5x5]
%onnx::Conv_769[FLOAT, 184x336x1x1]
%onnx::Conv_770[FLOAT, 184]
%onnx::Conv_772[FLOAT, 184x184x1x1]
%onnx::Conv_775[FLOAT, 184x1x5x5]
%onnx::Conv_778[FLOAT, 184x184x1x1]
%onnx::Conv_781[FLOAT, 184x92x1x1]
%onnx::Conv_784[FLOAT, 184x1x5x5]
%onnx::Conv_787[FLOAT, 184x92x1x1]
%onnx::Conv_790[FLOAT, 184x184x1x1]
%onnx::Conv_793[FLOAT, 184x1x3x3]
%onnx::Conv_796[FLOAT, 352x184x1x1]
%onnx::Conv_797[FLOAT, 352]
%onnx::Conv_799[FLOAT, 1504x352x1x1]
%onnx::Conv_800[FLOAT, 1504]
) {
%onnx::Conv_794 = Identity(%onnx::Conv_770)
%onnx::Conv_791 = Identity(%onnx::Conv_770)
%onnx::Conv_788 = Identity(%onnx::Conv_770)
%onnx::Conv_785 = Identity(%onnx::Conv_770)
%onnx::Conv_782 = Identity(%onnx::Conv_770)
%onnx::Conv_779 = Identity(%onnx::Conv_770)
%onnx::Conv_776 = Identity(%onnx::Conv_770)
%onnx::Conv_773 = Identity(%onnx::Conv_770)
%onnx::Conv_767 = Identity(%onnx::Conv_746)
%onnx::Conv_764 = Identity(%onnx::Conv_746)
%onnx::Conv_761 = Identity(%onnx::Conv_734)
%onnx::Conv_758 = Identity(%onnx::Conv_734)
%onnx::Conv_755 = Identity(%onnx::Conv_734)
%onnx::Conv_752 = Identity(%onnx::Conv_734)
%onnx::Conv_749 = Identity(%onnx::Conv_746)
%onnx::Conv_743 = Identity(%onnx::Conv_734)
%onnx::Conv_740 = Identity(%onnx::Conv_734)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_731 = Identity(%onnx::Conv_698)
%onnx::Conv_728 = Identity(%onnx::Conv_698)
%onnx::Conv_725 = Identity(%onnx::Conv_698)
%onnx::Conv_722 = Identity(%onnx::Conv_680)
%onnx::Conv_719 = Identity(%onnx::Conv_680)
%onnx::Conv_716 = Identity(%onnx::Conv_698)
%onnx::Conv_713 = Identity(%onnx::Conv_698)
%onnx::Conv_710 = Identity(%onnx::Conv_698)
%onnx::Conv_707 = Identity(%onnx::Conv_698)
%onnx::Conv_704 = Identity(%onnx::Conv_701)
%onnx::Conv_695 = Identity(%onnx::Conv_668)
%onnx::Conv_692 = Identity(%onnx::Conv_668)
%onnx::Conv_689 = Identity(%onnx::Conv_668)
%onnx::Conv_686 = Identity(%onnx::Conv_668)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_677 = Identity(%onnx::Conv_668)
%onnx::Conv_674 = Identity(%onnx::Conv_626)
%onnx::Conv_671 = Identity(%onnx::Conv_626)
%onnx::Conv_665 = Identity(%onnx::Conv_632)
%onnx::Conv_662 = Identity(%onnx::Conv_632)
%onnx::Conv_659 = Identity(%onnx::Conv_632)
%onnx::Conv_656 = Identity(%onnx::Conv_632)
%onnx::Conv_653 = Identity(%onnx::Conv_632)
%onnx::Conv_650 = Identity(%onnx::Conv_632)
%onnx::Conv_647 = Identity(%onnx::Conv_644)
%onnx::Conv_641 = Identity(%onnx::Conv_632)
%onnx::Conv_638 = Identity(%onnx::Conv_632)
%onnx::Conv_635 = Identity(%onnx::Conv_632)
%onnx::Conv_629 = Identity(%onnx::Conv_626)
%onnx::Conv_623 = Identity(%onnx::Conv_614)
%onnx::Conv_620 = Identity(%onnx::Conv_614)
%onnx::Conv_617 = Identity(%onnx::Conv_614)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_613, %onnx::Conv_614)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%611 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %611
}
|
val_accuracy
| 0 | 57,805,696 | 1,308,972 |
{'zcp_synflow': 83.53577055676055, 'zcp_zen': 68.34395599365234, 'zcp_epe_nas': 13.578428552432264, 'zcp_fisher': 0.15262392163276672, 'zcp_flops': 57805696.0, 'zcp_grad_norm': 24.772815704345703, 'zcp_grasp': -0.0221405029296875, 'zcp_jacov': -16.065438401281856, 'zcp_l2_norm': 578.9256591796875, 'zcp_nwot': 210.26772917280152, 'zcp_params': 1308972.0, 'zcp_plain': -0.006032541394233704, 'zcp_snip': 42.092838287353516, 'lat_1080ti_1': 0.5314426430197454, 'lat_1080ti_32': 0.5580937050419137, 'lat_1080ti_64': 0.4323053795162744, 'lat_2080ti_1': 0.6140306599782244, 'lat_2080ti_32': 0.5297712450080054, 'lat_2080ti_64': 0.4257330808983421, 'lat_essential_ph_1': 0.09433962264150944, 'lat_eyeriss': 0.33887737700108195, 'lat_fpga': 0.28650590249260327, 'lat_gold_6226': 0.24685788757146182, 'lat_gold_6240': 0.41937570911490024, 'lat_pixel2': 0.15217391304347827, 'lat_pixel3': 0.3540598829574173, 'lat_raspi4': 0.32155232844319814, 'lat_samsung_a50': 0.11578947368421053, 'lat_samsung_s7': 0.06299212598425197, 'lat_silver_4114': 0.4193008486845755, 'lat_silver_4210r': 0.43781293243275266, 'lat_titan_rtx_1': 0.5642199855130996, 'lat_titan_rtx_32': 0.5215788875842691, 'lat_titan_rtx_64': 0.43753770856775437, 'lat_titanx_1': 0.2973752598334911, 'lat_titanx_32': 0.48143985740581446, 'lat_titanx_64': 0.39214305320149606, 'lat_titanxp_1': 0.5401228364152573, 'lat_titanxp_32': 0.5166338695678039, 'lat_titanxp_64': 0.42297949266716706}
| |
FBNet_4688
|
FBNet
|
4688
|
4688
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_630[FLOAT, 16x3x3x3]
%onnx::Conv_631[FLOAT, 16]
%onnx::Conv_633[FLOAT, 96x16x1x1]
%onnx::Conv_634[FLOAT, 96]
%onnx::Conv_636[FLOAT, 96x1x3x3]
%onnx::Conv_639[FLOAT, 16x96x1x1]
%onnx::Conv_642[FLOAT, 96x16x1x1]
%onnx::Conv_645[FLOAT, 96x1x5x5]
%onnx::Conv_648[FLOAT, 24x96x1x1]
%onnx::Conv_649[FLOAT, 24]
%onnx::Conv_651[FLOAT, 72x24x1x1]
%onnx::Conv_652[FLOAT, 72]
%onnx::Conv_654[FLOAT, 72x1x3x3]
%onnx::Conv_657[FLOAT, 24x72x1x1]
%onnx::Conv_660[FLOAT, 24x12x1x1]
%onnx::Conv_663[FLOAT, 24x1x5x5]
%onnx::Conv_666[FLOAT, 24x12x1x1]
%onnx::Conv_669[FLOAT, 72x24x1x1]
%onnx::Conv_672[FLOAT, 72x1x3x3]
%onnx::Conv_675[FLOAT, 32x72x1x1]
%onnx::Conv_676[FLOAT, 32]
%onnx::Conv_678[FLOAT, 192x32x1x1]
%onnx::Conv_679[FLOAT, 192]
%onnx::Conv_681[FLOAT, 192x1x3x3]
%onnx::Conv_684[FLOAT, 32x192x1x1]
%onnx::Conv_687[FLOAT, 32x32x1x1]
%onnx::Conv_690[FLOAT, 32x1x5x5]
%onnx::Conv_693[FLOAT, 32x32x1x1]
%onnx::Conv_696[FLOAT, 192x32x1x1]
%onnx::Conv_699[FLOAT, 192x1x3x3]
%onnx::Conv_702[FLOAT, 32x192x1x1]
%onnx::Conv_705[FLOAT, 96x32x1x1]
%onnx::Conv_708[FLOAT, 96x1x5x5]
%onnx::Conv_711[FLOAT, 64x96x1x1]
%onnx::Conv_712[FLOAT, 64]
%onnx::Conv_714[FLOAT, 64x64x1x1]
%onnx::Conv_717[FLOAT, 64x1x5x5]
%onnx::Conv_720[FLOAT, 64x64x1x1]
%onnx::Conv_723[FLOAT, 192x64x1x1]
%onnx::Conv_726[FLOAT, 192x1x3x3]
%onnx::Conv_729[FLOAT, 64x192x1x1]
%onnx::Conv_732[FLOAT, 64x64x1x1]
%onnx::Conv_735[FLOAT, 64x1x5x5]
%onnx::Conv_738[FLOAT, 64x64x1x1]
%onnx::Conv_741[FLOAT, 192x64x1x1]
%onnx::Conv_744[FLOAT, 192x1x3x3]
%onnx::Conv_747[FLOAT, 112x192x1x1]
%onnx::Conv_748[FLOAT, 112]
%onnx::Conv_750[FLOAT, 672x112x1x1]
%onnx::Conv_751[FLOAT, 672]
%onnx::Conv_753[FLOAT, 672x1x3x3]
%onnx::Conv_756[FLOAT, 112x672x1x1]
%onnx::Conv_759[FLOAT, 112x112x1x1]
%onnx::Conv_762[FLOAT, 112x1x5x5]
%onnx::Conv_765[FLOAT, 112x112x1x1]
%onnx::Conv_768[FLOAT, 112x112x1x1]
%onnx::Conv_771[FLOAT, 112x1x3x3]
%onnx::Conv_774[FLOAT, 112x112x1x1]
%onnx::Conv_777[FLOAT, 112x56x1x1]
%onnx::Conv_780[FLOAT, 112x1x5x5]
%onnx::Conv_783[FLOAT, 184x56x1x1]
%onnx::Conv_784[FLOAT, 184]
%onnx::Conv_786[FLOAT, 184x184x1x1]
%onnx::Conv_789[FLOAT, 184x1x5x5]
%onnx::Conv_792[FLOAT, 184x184x1x1]
%onnx::Conv_795[FLOAT, 552x184x1x1]
%onnx::Conv_796[FLOAT, 552]
%onnx::Conv_798[FLOAT, 552x1x3x3]
%onnx::Conv_801[FLOAT, 184x552x1x1]
%onnx::Conv_804[FLOAT, 184x184x1x1]
%onnx::Conv_807[FLOAT, 184x1x3x3]
%onnx::Conv_810[FLOAT, 184x184x1x1]
%onnx::Conv_813[FLOAT, 1104x184x1x1]
%onnx::Conv_814[FLOAT, 1104]
%onnx::Conv_816[FLOAT, 1104x1x3x3]
%onnx::Conv_819[FLOAT, 352x1104x1x1]
%onnx::Conv_820[FLOAT, 352]
%onnx::Conv_822[FLOAT, 1504x352x1x1]
%onnx::Conv_823[FLOAT, 1504]
) {
%onnx::Conv_817 = Identity(%onnx::Conv_814)
%onnx::Conv_811 = Identity(%onnx::Conv_784)
%onnx::Conv_808 = Identity(%onnx::Conv_784)
%onnx::Conv_805 = Identity(%onnx::Conv_784)
%onnx::Conv_802 = Identity(%onnx::Conv_784)
%onnx::Conv_799 = Identity(%onnx::Conv_796)
%onnx::Conv_793 = Identity(%onnx::Conv_784)
%onnx::Conv_790 = Identity(%onnx::Conv_784)
%onnx::Conv_787 = Identity(%onnx::Conv_784)
%onnx::Conv_781 = Identity(%onnx::Conv_748)
%onnx::Conv_778 = Identity(%onnx::Conv_748)
%onnx::Conv_775 = Identity(%onnx::Conv_748)
%onnx::Conv_772 = Identity(%onnx::Conv_748)
%onnx::Conv_769 = Identity(%onnx::Conv_748)
%onnx::Conv_766 = Identity(%onnx::Conv_748)
%onnx::Conv_763 = Identity(%onnx::Conv_748)
%onnx::Conv_760 = Identity(%onnx::Conv_748)
%onnx::Conv_757 = Identity(%onnx::Conv_748)
%onnx::Conv_754 = Identity(%onnx::Conv_751)
%onnx::Conv_745 = Identity(%onnx::Conv_679)
%onnx::Conv_742 = Identity(%onnx::Conv_679)
%onnx::Conv_739 = Identity(%onnx::Conv_712)
%onnx::Conv_736 = Identity(%onnx::Conv_712)
%onnx::Conv_733 = Identity(%onnx::Conv_712)
%onnx::Conv_730 = Identity(%onnx::Conv_712)
%onnx::Conv_727 = Identity(%onnx::Conv_679)
%onnx::Conv_724 = Identity(%onnx::Conv_679)
%onnx::Conv_721 = Identity(%onnx::Conv_712)
%onnx::Conv_718 = Identity(%onnx::Conv_712)
%onnx::Conv_715 = Identity(%onnx::Conv_712)
%onnx::Conv_709 = Identity(%onnx::Conv_634)
%onnx::Conv_706 = Identity(%onnx::Conv_634)
%onnx::Conv_703 = Identity(%onnx::Conv_676)
%onnx::Conv_700 = Identity(%onnx::Conv_679)
%onnx::Conv_697 = Identity(%onnx::Conv_679)
%onnx::Conv_694 = Identity(%onnx::Conv_676)
%onnx::Conv_691 = Identity(%onnx::Conv_676)
%onnx::Conv_688 = Identity(%onnx::Conv_676)
%onnx::Conv_685 = Identity(%onnx::Conv_676)
%onnx::Conv_682 = Identity(%onnx::Conv_679)
%onnx::Conv_673 = Identity(%onnx::Conv_652)
%onnx::Conv_670 = Identity(%onnx::Conv_652)
%onnx::Conv_667 = Identity(%onnx::Conv_649)
%onnx::Conv_664 = Identity(%onnx::Conv_649)
%onnx::Conv_661 = Identity(%onnx::Conv_649)
%onnx::Conv_658 = Identity(%onnx::Conv_649)
%onnx::Conv_655 = Identity(%onnx::Conv_652)
%onnx::Conv_646 = Identity(%onnx::Conv_634)
%onnx::Conv_643 = Identity(%onnx::Conv_634)
%onnx::Conv_640 = Identity(%onnx::Conv_631)
%onnx::Conv_637 = Identity(%onnx::Conv_634)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_630, %onnx::Conv_631)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%628 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %628
}
|
val_accuracy
| 0 | 74,600,064 | 2,030,500 |
{'zcp_synflow': 83.24248826225511, 'zcp_zen': 73.03765106201172, 'zcp_epe_nas': 20.86995528682622, 'zcp_fisher': 0.21869702637195587, 'zcp_flops': 74600064.0, 'zcp_grad_norm': 28.20191192626953, 'zcp_grasp': -0.20346450805664062, 'zcp_jacov': -16.048911329834308, 'zcp_l2_norm': 673.8461303710938, 'zcp_nwot': 214.88971802505992, 'zcp_params': 2030500.0, 'zcp_plain': 0.01106171403080225, 'zcp_snip': 53.12782287597656, 'lat_1080ti_1': 0.6673204579797035, 'lat_1080ti_32': 0.5775897045177855, 'lat_1080ti_64': 0.5030368363002787, 'lat_2080ti_1': 0.6908351644953499, 'lat_2080ti_32': 0.6117358459072542, 'lat_2080ti_64': 0.5566656344582124, 'lat_essential_ph_1': 0.33962264150943394, 'lat_eyeriss': 0.5282856252177581, 'lat_fpga': 0.5738924786510851, 'lat_gold_6226': 0.3486014346318307, 'lat_gold_6240': 0.5267728124831632, 'lat_pixel2': 0.32608695652173914, 'lat_pixel3': 0.47039266909308275, 'lat_raspi4': 0.5115815083004254, 'lat_samsung_a50': 0.21052631578947367, 'lat_samsung_s7': 0.18110236220472442, 'lat_silver_4114': 0.538597960944417, 'lat_silver_4210r': 0.5531759166307262, 'lat_titan_rtx_1': 0.6281794590540382, 'lat_titan_rtx_32': 0.5908494683455736, 'lat_titan_rtx_64': 0.5423854598742222, 'lat_titanx_1': 0.33709596760391725, 'lat_titanx_32': 0.5398508477788194, 'lat_titanx_64': 0.544315842732137, 'lat_titanxp_1': 0.5984552444329821, 'lat_titanxp_32': 0.5963076544718627, 'lat_titanxp_64': 0.5100114793139011}
| |
FBNet_415
|
FBNet
|
415
|
415
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_652[FLOAT, 16x3x3x3]
%onnx::Conv_653[FLOAT, 16]
%onnx::Conv_655[FLOAT, 16x8x1x1]
%onnx::Conv_658[FLOAT, 16x1x5x5]
%onnx::Conv_661[FLOAT, 16x8x1x1]
%onnx::Conv_664[FLOAT, 16x8x1x1]
%onnx::Conv_667[FLOAT, 16x1x5x5]
%onnx::Conv_670[FLOAT, 24x8x1x1]
%onnx::Conv_671[FLOAT, 24]
%onnx::Conv_673[FLOAT, 72x24x1x1]
%onnx::Conv_674[FLOAT, 72]
%onnx::Conv_676[FLOAT, 72x1x5x5]
%onnx::Conv_679[FLOAT, 24x72x1x1]
%onnx::Conv_682[FLOAT, 24x24x1x1]
%onnx::Conv_685[FLOAT, 24x1x5x5]
%onnx::Conv_688[FLOAT, 24x24x1x1]
%onnx::Conv_691[FLOAT, 24x24x1x1]
%onnx::Conv_694[FLOAT, 24x1x5x5]
%onnx::Conv_697[FLOAT, 24x24x1x1]
%onnx::Conv_700[FLOAT, 32x24x1x1]
%onnx::Conv_701[FLOAT, 32]
%onnx::Conv_703[FLOAT, 192x32x1x1]
%onnx::Conv_704[FLOAT, 192]
%onnx::Conv_706[FLOAT, 192x1x3x3]
%onnx::Conv_709[FLOAT, 32x192x1x1]
%onnx::Conv_712[FLOAT, 32x32x1x1]
%onnx::Conv_715[FLOAT, 32x1x5x5]
%onnx::Conv_718[FLOAT, 32x32x1x1]
%onnx::Conv_721[FLOAT, 32x16x1x1]
%onnx::Conv_724[FLOAT, 32x1x5x5]
%onnx::Conv_727[FLOAT, 64x16x1x1]
%onnx::Conv_728[FLOAT, 64]
%onnx::Conv_730[FLOAT, 64x32x1x1]
%onnx::Conv_733[FLOAT, 64x1x3x3]
%onnx::Conv_736[FLOAT, 64x32x1x1]
%onnx::Conv_739[FLOAT, 192x64x1x1]
%onnx::Conv_742[FLOAT, 192x1x5x5]
%onnx::Conv_745[FLOAT, 64x192x1x1]
%onnx::Conv_748[FLOAT, 64x64x1x1]
%onnx::Conv_751[FLOAT, 64x1x5x5]
%onnx::Conv_754[FLOAT, 64x64x1x1]
%onnx::Conv_757[FLOAT, 384x64x1x1]
%onnx::Conv_758[FLOAT, 384]
%onnx::Conv_760[FLOAT, 384x1x3x3]
%onnx::Conv_763[FLOAT, 112x384x1x1]
%onnx::Conv_764[FLOAT, 112]
%onnx::Conv_766[FLOAT, 336x112x1x1]
%onnx::Conv_767[FLOAT, 336]
%onnx::Conv_769[FLOAT, 336x1x3x3]
%onnx::Conv_772[FLOAT, 112x336x1x1]
%onnx::Conv_775[FLOAT, 336x112x1x1]
%onnx::Conv_778[FLOAT, 336x1x5x5]
%onnx::Conv_781[FLOAT, 112x336x1x1]
%onnx::Conv_784[FLOAT, 672x112x1x1]
%onnx::Conv_785[FLOAT, 672]
%onnx::Conv_787[FLOAT, 672x1x3x3]
%onnx::Conv_790[FLOAT, 112x672x1x1]
%onnx::Conv_793[FLOAT, 184x112x1x1]
%onnx::Conv_794[FLOAT, 184]
%onnx::Conv_796[FLOAT, 184x184x1x1]
%onnx::Conv_799[FLOAT, 184x1x3x3]
%onnx::Conv_802[FLOAT, 184x184x1x1]
%onnx::Conv_805[FLOAT, 184x184x1x1]
%onnx::Conv_808[FLOAT, 184x1x3x3]
%onnx::Conv_811[FLOAT, 184x184x1x1]
%onnx::Conv_814[FLOAT, 1104x184x1x1]
%onnx::Conv_815[FLOAT, 1104]
%onnx::Conv_817[FLOAT, 1104x1x5x5]
%onnx::Conv_820[FLOAT, 184x1104x1x1]
%onnx::Conv_823[FLOAT, 184x92x1x1]
%onnx::Conv_826[FLOAT, 184x1x5x5]
%onnx::Conv_829[FLOAT, 352x92x1x1]
%onnx::Conv_830[FLOAT, 352]
%onnx::Conv_832[FLOAT, 1504x352x1x1]
%onnx::Conv_833[FLOAT, 1504]
) {
%onnx::Conv_827 = Identity(%onnx::Conv_794)
%onnx::Conv_824 = Identity(%onnx::Conv_794)
%onnx::Conv_821 = Identity(%onnx::Conv_794)
%onnx::Conv_818 = Identity(%onnx::Conv_815)
%onnx::Conv_812 = Identity(%onnx::Conv_794)
%onnx::Conv_809 = Identity(%onnx::Conv_794)
%onnx::Conv_806 = Identity(%onnx::Conv_794)
%onnx::Conv_803 = Identity(%onnx::Conv_794)
%onnx::Conv_800 = Identity(%onnx::Conv_794)
%onnx::Conv_797 = Identity(%onnx::Conv_794)
%onnx::Conv_791 = Identity(%onnx::Conv_764)
%onnx::Conv_788 = Identity(%onnx::Conv_785)
%onnx::Conv_782 = Identity(%onnx::Conv_764)
%onnx::Conv_779 = Identity(%onnx::Conv_767)
%onnx::Conv_776 = Identity(%onnx::Conv_767)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_767)
%onnx::Conv_761 = Identity(%onnx::Conv_758)
%onnx::Conv_755 = Identity(%onnx::Conv_728)
%onnx::Conv_752 = Identity(%onnx::Conv_728)
%onnx::Conv_749 = Identity(%onnx::Conv_728)
%onnx::Conv_746 = Identity(%onnx::Conv_728)
%onnx::Conv_743 = Identity(%onnx::Conv_704)
%onnx::Conv_740 = Identity(%onnx::Conv_704)
%onnx::Conv_737 = Identity(%onnx::Conv_728)
%onnx::Conv_734 = Identity(%onnx::Conv_728)
%onnx::Conv_731 = Identity(%onnx::Conv_728)
%onnx::Conv_725 = Identity(%onnx::Conv_701)
%onnx::Conv_722 = Identity(%onnx::Conv_701)
%onnx::Conv_719 = Identity(%onnx::Conv_701)
%onnx::Conv_716 = Identity(%onnx::Conv_701)
%onnx::Conv_713 = Identity(%onnx::Conv_701)
%onnx::Conv_710 = Identity(%onnx::Conv_701)
%onnx::Conv_707 = Identity(%onnx::Conv_704)
%onnx::Conv_698 = Identity(%onnx::Conv_671)
%onnx::Conv_695 = Identity(%onnx::Conv_671)
%onnx::Conv_692 = Identity(%onnx::Conv_671)
%onnx::Conv_689 = Identity(%onnx::Conv_671)
%onnx::Conv_686 = Identity(%onnx::Conv_671)
%onnx::Conv_683 = Identity(%onnx::Conv_671)
%onnx::Conv_680 = Identity(%onnx::Conv_671)
%onnx::Conv_677 = Identity(%onnx::Conv_674)
%onnx::Conv_668 = Identity(%onnx::Conv_653)
%onnx::Conv_665 = Identity(%onnx::Conv_653)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_652, %onnx::Conv_653)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.4/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%650 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %650
}
|
val_accuracy
| 0 | 64,731,648 | 1,814,468 |
{'zcp_synflow': 77.49783375402701, 'zcp_zen': 66.55447387695312, 'zcp_epe_nas': 24.3980467893341, 'zcp_fisher': 0.07563282549381256, 'zcp_flops': 64731648.0, 'zcp_grad_norm': 21.768558502197266, 'zcp_grasp': -0.06540966033935547, 'zcp_jacov': -16.067813768417636, 'zcp_l2_norm': 608.1273193359375, 'zcp_nwot': 208.03672307903614, 'zcp_params': 1814468.0, 'zcp_plain': 0.009970569051802158, 'zcp_snip': 40.44108963012695, 'lat_1080ti_1': 0.5122999799316379, 'lat_1080ti_32': 0.43309113967095575, 'lat_1080ti_64': 0.3274538304103022, 'lat_2080ti_1': 0.5709215678533135, 'lat_2080ti_32': 0.4602956821975326, 'lat_2080ti_64': 0.32173228902895373, 'lat_essential_ph_1': 0.22641509433962265, 'lat_eyeriss': 0.3586085489520106, 'lat_fpga': 0.4555853511315172, 'lat_gold_6226': 0.31887074918330277, 'lat_gold_6240': 0.49354947753674594, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.38076116640623253, 'lat_raspi4': 0.3931196165239907, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.2047244094488189, 'lat_silver_4114': 0.5052979251697649, 'lat_silver_4210r': 0.5170568753268461, 'lat_titan_rtx_1': 0.5542737051765205, 'lat_titan_rtx_32': 0.47863711639959355, 'lat_titan_rtx_64': 0.34723434581337875, 'lat_titanx_1': 0.2975579348041228, 'lat_titanx_32': 0.3806106874770237, 'lat_titanx_64': 0.3090731424106403, 'lat_titanxp_1': 0.5337579450239193, 'lat_titanxp_32': 0.43922131047335555, 'lat_titanxp_64': 0.3448879457393271}
| |
FBNet_3536
|
FBNet
|
3536
|
3536
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_613[FLOAT, 16x3x3x3]
%onnx::Conv_614[FLOAT, 16]
%onnx::Conv_616[FLOAT, 48x16x1x1]
%onnx::Conv_617[FLOAT, 48]
%onnx::Conv_619[FLOAT, 48x1x3x3]
%onnx::Conv_622[FLOAT, 24x48x1x1]
%onnx::Conv_623[FLOAT, 24]
%onnx::Conv_625[FLOAT, 24x12x1x1]
%onnx::Conv_628[FLOAT, 24x1x5x5]
%onnx::Conv_631[FLOAT, 24x12x1x1]
%onnx::Conv_634[FLOAT, 24x24x1x1]
%onnx::Conv_637[FLOAT, 24x1x3x3]
%onnx::Conv_640[FLOAT, 24x24x1x1]
%onnx::Conv_643[FLOAT, 24x24x1x1]
%onnx::Conv_646[FLOAT, 24x1x5x5]
%onnx::Conv_649[FLOAT, 32x24x1x1]
%onnx::Conv_650[FLOAT, 32]
%onnx::Conv_652[FLOAT, 96x32x1x1]
%onnx::Conv_653[FLOAT, 96]
%onnx::Conv_655[FLOAT, 96x1x3x3]
%onnx::Conv_658[FLOAT, 32x96x1x1]
%onnx::Conv_661[FLOAT, 96x32x1x1]
%onnx::Conv_664[FLOAT, 96x1x3x3]
%onnx::Conv_667[FLOAT, 32x96x1x1]
%onnx::Conv_670[FLOAT, 32x32x1x1]
%onnx::Conv_673[FLOAT, 32x1x3x3]
%onnx::Conv_676[FLOAT, 32x32x1x1]
%onnx::Conv_679[FLOAT, 32x32x1x1]
%onnx::Conv_682[FLOAT, 32x1x5x5]
%onnx::Conv_685[FLOAT, 64x32x1x1]
%onnx::Conv_686[FLOAT, 64]
%onnx::Conv_688[FLOAT, 64x32x1x1]
%onnx::Conv_691[FLOAT, 64x1x5x5]
%onnx::Conv_694[FLOAT, 64x32x1x1]
%onnx::Conv_697[FLOAT, 64x32x1x1]
%onnx::Conv_700[FLOAT, 64x1x3x3]
%onnx::Conv_703[FLOAT, 64x32x1x1]
%onnx::Conv_706[FLOAT, 384x64x1x1]
%onnx::Conv_707[FLOAT, 384]
%onnx::Conv_709[FLOAT, 384x1x3x3]
%onnx::Conv_712[FLOAT, 64x384x1x1]
%onnx::Conv_715[FLOAT, 64x64x1x1]
%onnx::Conv_718[FLOAT, 64x1x5x5]
%onnx::Conv_721[FLOAT, 112x64x1x1]
%onnx::Conv_722[FLOAT, 112]
%onnx::Conv_724[FLOAT, 336x112x1x1]
%onnx::Conv_725[FLOAT, 336]
%onnx::Conv_727[FLOAT, 336x1x5x5]
%onnx::Conv_730[FLOAT, 112x336x1x1]
%onnx::Conv_733[FLOAT, 672x112x1x1]
%onnx::Conv_734[FLOAT, 672]
%onnx::Conv_736[FLOAT, 672x1x3x3]
%onnx::Conv_739[FLOAT, 112x672x1x1]
%onnx::Conv_742[FLOAT, 336x112x1x1]
%onnx::Conv_745[FLOAT, 336x1x3x3]
%onnx::Conv_748[FLOAT, 184x336x1x1]
%onnx::Conv_749[FLOAT, 184]
%onnx::Conv_751[FLOAT, 184x92x1x1]
%onnx::Conv_754[FLOAT, 184x1x3x3]
%onnx::Conv_757[FLOAT, 184x92x1x1]
%onnx::Conv_760[FLOAT, 552x184x1x1]
%onnx::Conv_761[FLOAT, 552]
%onnx::Conv_763[FLOAT, 552x1x3x3]
%onnx::Conv_766[FLOAT, 184x552x1x1]
%onnx::Conv_769[FLOAT, 552x184x1x1]
%onnx::Conv_772[FLOAT, 552x1x5x5]
%onnx::Conv_775[FLOAT, 184x552x1x1]
%onnx::Conv_778[FLOAT, 552x184x1x1]
%onnx::Conv_781[FLOAT, 552x1x5x5]
%onnx::Conv_784[FLOAT, 352x552x1x1]
%onnx::Conv_785[FLOAT, 352]
%onnx::Conv_787[FLOAT, 1504x352x1x1]
%onnx::Conv_788[FLOAT, 1504]
) {
%onnx::Conv_782 = Identity(%onnx::Conv_761)
%onnx::Conv_779 = Identity(%onnx::Conv_761)
%onnx::Conv_776 = Identity(%onnx::Conv_749)
%onnx::Conv_773 = Identity(%onnx::Conv_761)
%onnx::Conv_770 = Identity(%onnx::Conv_761)
%onnx::Conv_767 = Identity(%onnx::Conv_749)
%onnx::Conv_764 = Identity(%onnx::Conv_761)
%onnx::Conv_758 = Identity(%onnx::Conv_749)
%onnx::Conv_755 = Identity(%onnx::Conv_749)
%onnx::Conv_752 = Identity(%onnx::Conv_749)
%onnx::Conv_746 = Identity(%onnx::Conv_725)
%onnx::Conv_743 = Identity(%onnx::Conv_725)
%onnx::Conv_740 = Identity(%onnx::Conv_722)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_731 = Identity(%onnx::Conv_722)
%onnx::Conv_728 = Identity(%onnx::Conv_725)
%onnx::Conv_719 = Identity(%onnx::Conv_686)
%onnx::Conv_716 = Identity(%onnx::Conv_686)
%onnx::Conv_713 = Identity(%onnx::Conv_686)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_686)
%onnx::Conv_701 = Identity(%onnx::Conv_686)
%onnx::Conv_698 = Identity(%onnx::Conv_686)
%onnx::Conv_695 = Identity(%onnx::Conv_686)
%onnx::Conv_692 = Identity(%onnx::Conv_686)
%onnx::Conv_689 = Identity(%onnx::Conv_686)
%onnx::Conv_683 = Identity(%onnx::Conv_650)
%onnx::Conv_680 = Identity(%onnx::Conv_650)
%onnx::Conv_677 = Identity(%onnx::Conv_650)
%onnx::Conv_674 = Identity(%onnx::Conv_650)
%onnx::Conv_671 = Identity(%onnx::Conv_650)
%onnx::Conv_668 = Identity(%onnx::Conv_650)
%onnx::Conv_665 = Identity(%onnx::Conv_653)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_650)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_647 = Identity(%onnx::Conv_623)
%onnx::Conv_644 = Identity(%onnx::Conv_623)
%onnx::Conv_641 = Identity(%onnx::Conv_623)
%onnx::Conv_638 = Identity(%onnx::Conv_623)
%onnx::Conv_635 = Identity(%onnx::Conv_623)
%onnx::Conv_632 = Identity(%onnx::Conv_623)
%onnx::Conv_629 = Identity(%onnx::Conv_623)
%onnx::Conv_626 = Identity(%onnx::Conv_623)
%onnx::Conv_620 = Identity(%onnx::Conv_617)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_613, %onnx::Conv_614)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%611 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %611
}
|
val_accuracy
| 0 | 57,696,640 | 1,919,532 |
{'zcp_synflow': 73.05816576522047, 'zcp_zen': 64.99130249023438, 'zcp_epe_nas': 10.075197277893766, 'zcp_fisher': 0.08555657416582108, 'zcp_flops': 57696640.0, 'zcp_grad_norm': 16.870935440063477, 'zcp_grasp': 0.03309822082519531, 'zcp_jacov': -16.050492177568323, 'zcp_l2_norm': 594.5985717773438, 'zcp_nwot': 205.23244376962504, 'zcp_params': 1919532.0, 'zcp_plain': -0.0031644566915929317, 'zcp_snip': 32.743160247802734, 'lat_1080ti_1': 0.4635131974204375, 'lat_1080ti_32': 0.346969440730971, 'lat_1080ti_64': 0.18487492927445975, 'lat_2080ti_1': 0.5008996660139856, 'lat_2080ti_32': 0.36150242582425623, 'lat_2080ti_64': 0.19707192312619878, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.25446977059761966, 'lat_fpga': 0.36285734076133735, 'lat_gold_6226': 0.3347998376206203, 'lat_gold_6240': 0.5373282745554907, 'lat_pixel2': 0.21739130434782608, 'lat_pixel3': 0.24637854006223536, 'lat_raspi4': 0.29525930799616795, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.13385826771653545, 'lat_silver_4114': 0.4888598888210124, 'lat_silver_4210r': 0.49718581701191056, 'lat_titan_rtx_1': 0.464637281163517, 'lat_titan_rtx_32': 0.3707380823155452, 'lat_titan_rtx_64': 0.22023519280466244, 'lat_titanx_1': 0.2647764781451052, 'lat_titanx_32': 0.26182597959164045, 'lat_titanx_64': 0.1640970773600223, 'lat_titanxp_1': 0.4457287200684146, 'lat_titanxp_32': 0.3219881551899596, 'lat_titanxp_64': 0.19773244417438138}
| |
FBNet_651
|
FBNet
|
651
|
651
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_669[FLOAT, 16x3x3x3]
%onnx::Conv_670[FLOAT, 16]
%onnx::Conv_672[FLOAT, 48x16x1x1]
%onnx::Conv_673[FLOAT, 48]
%onnx::Conv_675[FLOAT, 48x1x3x3]
%onnx::Conv_678[FLOAT, 24x48x1x1]
%onnx::Conv_679[FLOAT, 24]
%onnx::Conv_681[FLOAT, 24x24x1x1]
%onnx::Conv_684[FLOAT, 24x1x5x5]
%onnx::Conv_687[FLOAT, 24x24x1x1]
%onnx::Conv_690[FLOAT, 24x12x1x1]
%onnx::Conv_693[FLOAT, 24x1x3x3]
%onnx::Conv_696[FLOAT, 24x12x1x1]
%onnx::Conv_699[FLOAT, 24x24x1x1]
%onnx::Conv_702[FLOAT, 24x1x5x5]
%onnx::Conv_705[FLOAT, 24x24x1x1]
%onnx::Conv_708[FLOAT, 72x24x1x1]
%onnx::Conv_709[FLOAT, 72]
%onnx::Conv_711[FLOAT, 72x1x3x3]
%onnx::Conv_714[FLOAT, 32x72x1x1]
%onnx::Conv_715[FLOAT, 32]
%onnx::Conv_717[FLOAT, 32x32x1x1]
%onnx::Conv_720[FLOAT, 32x1x3x3]
%onnx::Conv_723[FLOAT, 32x32x1x1]
%onnx::Conv_726[FLOAT, 32x16x1x1]
%onnx::Conv_729[FLOAT, 32x1x5x5]
%onnx::Conv_732[FLOAT, 32x16x1x1]
%onnx::Conv_735[FLOAT, 96x32x1x1]
%onnx::Conv_736[FLOAT, 96]
%onnx::Conv_738[FLOAT, 96x1x5x5]
%onnx::Conv_741[FLOAT, 32x96x1x1]
%onnx::Conv_744[FLOAT, 32x32x1x1]
%onnx::Conv_747[FLOAT, 32x1x3x3]
%onnx::Conv_750[FLOAT, 64x32x1x1]
%onnx::Conv_751[FLOAT, 64]
%onnx::Conv_753[FLOAT, 192x64x1x1]
%onnx::Conv_754[FLOAT, 192]
%onnx::Conv_756[FLOAT, 192x1x5x5]
%onnx::Conv_759[FLOAT, 64x192x1x1]
%onnx::Conv_762[FLOAT, 64x64x1x1]
%onnx::Conv_765[FLOAT, 64x1x3x3]
%onnx::Conv_768[FLOAT, 64x64x1x1]
%onnx::Conv_771[FLOAT, 64x32x1x1]
%onnx::Conv_774[FLOAT, 64x1x5x5]
%onnx::Conv_777[FLOAT, 64x32x1x1]
%onnx::Conv_780[FLOAT, 64x32x1x1]
%onnx::Conv_783[FLOAT, 64x1x3x3]
%onnx::Conv_786[FLOAT, 112x32x1x1]
%onnx::Conv_787[FLOAT, 112]
%onnx::Conv_789[FLOAT, 112x56x1x1]
%onnx::Conv_792[FLOAT, 112x1x5x5]
%onnx::Conv_795[FLOAT, 112x56x1x1]
%onnx::Conv_798[FLOAT, 112x112x1x1]
%onnx::Conv_801[FLOAT, 112x1x3x3]
%onnx::Conv_804[FLOAT, 112x112x1x1]
%onnx::Conv_807[FLOAT, 112x112x1x1]
%onnx::Conv_810[FLOAT, 112x1x5x5]
%onnx::Conv_813[FLOAT, 184x112x1x1]
%onnx::Conv_814[FLOAT, 184]
%onnx::Conv_816[FLOAT, 184x92x1x1]
%onnx::Conv_819[FLOAT, 184x1x3x3]
%onnx::Conv_822[FLOAT, 184x92x1x1]
%onnx::Conv_825[FLOAT, 184x184x1x1]
%onnx::Conv_828[FLOAT, 184x1x5x5]
%onnx::Conv_831[FLOAT, 184x184x1x1]
%onnx::Conv_834[FLOAT, 184x92x1x1]
%onnx::Conv_837[FLOAT, 184x1x3x3]
%onnx::Conv_840[FLOAT, 352x92x1x1]
%onnx::Conv_841[FLOAT, 352]
%onnx::Conv_843[FLOAT, 1504x352x1x1]
%onnx::Conv_844[FLOAT, 1504]
) {
%onnx::Conv_838 = Identity(%onnx::Conv_814)
%onnx::Conv_835 = Identity(%onnx::Conv_814)
%onnx::Conv_832 = Identity(%onnx::Conv_814)
%onnx::Conv_829 = Identity(%onnx::Conv_814)
%onnx::Conv_826 = Identity(%onnx::Conv_814)
%onnx::Conv_823 = Identity(%onnx::Conv_814)
%onnx::Conv_820 = Identity(%onnx::Conv_814)
%onnx::Conv_817 = Identity(%onnx::Conv_814)
%onnx::Conv_811 = Identity(%onnx::Conv_787)
%onnx::Conv_808 = Identity(%onnx::Conv_787)
%onnx::Conv_805 = Identity(%onnx::Conv_787)
%onnx::Conv_802 = Identity(%onnx::Conv_787)
%onnx::Conv_799 = Identity(%onnx::Conv_787)
%onnx::Conv_796 = Identity(%onnx::Conv_787)
%onnx::Conv_793 = Identity(%onnx::Conv_787)
%onnx::Conv_790 = Identity(%onnx::Conv_787)
%onnx::Conv_784 = Identity(%onnx::Conv_751)
%onnx::Conv_781 = Identity(%onnx::Conv_751)
%onnx::Conv_778 = Identity(%onnx::Conv_751)
%onnx::Conv_775 = Identity(%onnx::Conv_751)
%onnx::Conv_772 = Identity(%onnx::Conv_751)
%onnx::Conv_769 = Identity(%onnx::Conv_751)
%onnx::Conv_766 = Identity(%onnx::Conv_751)
%onnx::Conv_763 = Identity(%onnx::Conv_751)
%onnx::Conv_760 = Identity(%onnx::Conv_751)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_748 = Identity(%onnx::Conv_715)
%onnx::Conv_745 = Identity(%onnx::Conv_715)
%onnx::Conv_742 = Identity(%onnx::Conv_715)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_733 = Identity(%onnx::Conv_715)
%onnx::Conv_730 = Identity(%onnx::Conv_715)
%onnx::Conv_727 = Identity(%onnx::Conv_715)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_715)
%onnx::Conv_718 = Identity(%onnx::Conv_715)
%onnx::Conv_712 = Identity(%onnx::Conv_709)
%onnx::Conv_706 = Identity(%onnx::Conv_679)
%onnx::Conv_703 = Identity(%onnx::Conv_679)
%onnx::Conv_700 = Identity(%onnx::Conv_679)
%onnx::Conv_697 = Identity(%onnx::Conv_679)
%onnx::Conv_694 = Identity(%onnx::Conv_679)
%onnx::Conv_691 = Identity(%onnx::Conv_679)
%onnx::Conv_688 = Identity(%onnx::Conv_679)
%onnx::Conv_685 = Identity(%onnx::Conv_679)
%onnx::Conv_682 = Identity(%onnx::Conv_679)
%onnx::Conv_676 = Identity(%onnx::Conv_673)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_669, %onnx::Conv_670)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%667 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %667
}
|
val_accuracy
| 0 | 33,448,320 | 1,007,068 |
{'zcp_synflow': 68.78046910241163, 'zcp_zen': 57.479095458984375, 'zcp_epe_nas': 7.852800604797541, 'zcp_fisher': 0.06484105437994003, 'zcp_flops': 33448320.0, 'zcp_grad_norm': 17.552143096923828, 'zcp_grasp': -0.08158302307128906, 'zcp_jacov': -16.070682762007195, 'zcp_l2_norm': 453.0606994628906, 'zcp_nwot': 202.68656109424163, 'zcp_params': 1007068.0, 'zcp_plain': 0.00010264058073516935, 'zcp_snip': 29.901294708251953, 'lat_1080ti_1': 0.48034217561152165, 'lat_1080ti_32': 0.4267432801946257, 'lat_1080ti_64': 0.2110201833445552, 'lat_2080ti_1': 0.5357167299275604, 'lat_2080ti_32': 0.3967714646448104, 'lat_2080ti_64': 0.2357145700874188, 'lat_essential_ph_1': 0.07547169811320754, 'lat_eyeriss': 0.030504465186217487, 'lat_fpga': 0.0017811377512146054, 'lat_gold_6226': 0.0, 'lat_gold_6240': 0.1887234493761968, 'lat_pixel2': 0.021739130434782608, 'lat_pixel3': 0.061423090501911214, 'lat_raspi4': 0.08972093794486691, 'lat_samsung_a50': 0.010526315789473684, 'lat_samsung_s7': 0.023622047244094488, 'lat_silver_4114': 0.37450406549936804, 'lat_silver_4210r': 0.29460613686206205, 'lat_titan_rtx_1': 0.5131530974223125, 'lat_titan_rtx_32': 0.41660002998620244, 'lat_titan_rtx_64': 0.25330007035991997, 'lat_titanx_1': 0.27040384239144416, 'lat_titanx_32': 0.2923830079450712, 'lat_titanx_64': 0.2075909759729978, 'lat_titanxp_1': 0.48884851490075304, 'lat_titanxp_32': 0.3680061619805603, 'lat_titanxp_64': 0.23686060372443687}
| |
FBNet_1806
|
FBNet
|
1806
|
1806
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_642[FLOAT, 16x3x3x3]
%onnx::Conv_643[FLOAT, 16]
%onnx::Conv_645[FLOAT, 16x8x1x1]
%onnx::Conv_648[FLOAT, 16x1x5x5]
%onnx::Conv_651[FLOAT, 16x8x1x1]
%onnx::Conv_654[FLOAT, 16x16x1x1]
%onnx::Conv_657[FLOAT, 16x1x5x5]
%onnx::Conv_660[FLOAT, 24x16x1x1]
%onnx::Conv_661[FLOAT, 24]
%onnx::Conv_663[FLOAT, 144x24x1x1]
%onnx::Conv_664[FLOAT, 144]
%onnx::Conv_666[FLOAT, 144x1x3x3]
%onnx::Conv_669[FLOAT, 24x144x1x1]
%onnx::Conv_672[FLOAT, 24x24x1x1]
%onnx::Conv_675[FLOAT, 24x1x3x3]
%onnx::Conv_678[FLOAT, 24x24x1x1]
%onnx::Conv_681[FLOAT, 24x24x1x1]
%onnx::Conv_684[FLOAT, 24x1x5x5]
%onnx::Conv_687[FLOAT, 24x24x1x1]
%onnx::Conv_690[FLOAT, 32x24x1x1]
%onnx::Conv_691[FLOAT, 32]
%onnx::Conv_693[FLOAT, 32x32x1x1]
%onnx::Conv_696[FLOAT, 32x1x3x3]
%onnx::Conv_699[FLOAT, 32x32x1x1]
%onnx::Conv_702[FLOAT, 192x32x1x1]
%onnx::Conv_703[FLOAT, 192]
%onnx::Conv_705[FLOAT, 192x1x5x5]
%onnx::Conv_708[FLOAT, 32x192x1x1]
%onnx::Conv_711[FLOAT, 96x32x1x1]
%onnx::Conv_712[FLOAT, 96]
%onnx::Conv_714[FLOAT, 96x1x3x3]
%onnx::Conv_717[FLOAT, 64x96x1x1]
%onnx::Conv_718[FLOAT, 64]
%onnx::Conv_720[FLOAT, 64x64x1x1]
%onnx::Conv_723[FLOAT, 64x1x5x5]
%onnx::Conv_726[FLOAT, 64x64x1x1]
%onnx::Conv_729[FLOAT, 64x32x1x1]
%onnx::Conv_732[FLOAT, 64x1x3x3]
%onnx::Conv_735[FLOAT, 64x32x1x1]
%onnx::Conv_738[FLOAT, 192x64x1x1]
%onnx::Conv_741[FLOAT, 192x1x3x3]
%onnx::Conv_744[FLOAT, 64x192x1x1]
%onnx::Conv_747[FLOAT, 192x64x1x1]
%onnx::Conv_750[FLOAT, 192x1x3x3]
%onnx::Conv_753[FLOAT, 112x192x1x1]
%onnx::Conv_754[FLOAT, 112]
%onnx::Conv_756[FLOAT, 672x112x1x1]
%onnx::Conv_757[FLOAT, 672]
%onnx::Conv_759[FLOAT, 672x1x3x3]
%onnx::Conv_762[FLOAT, 112x672x1x1]
%onnx::Conv_765[FLOAT, 112x112x1x1]
%onnx::Conv_768[FLOAT, 112x1x5x5]
%onnx::Conv_771[FLOAT, 112x112x1x1]
%onnx::Conv_774[FLOAT, 672x112x1x1]
%onnx::Conv_777[FLOAT, 672x1x5x5]
%onnx::Conv_780[FLOAT, 112x672x1x1]
%onnx::Conv_783[FLOAT, 112x56x1x1]
%onnx::Conv_786[FLOAT, 112x1x5x5]
%onnx::Conv_789[FLOAT, 184x56x1x1]
%onnx::Conv_790[FLOAT, 184]
%onnx::Conv_792[FLOAT, 184x92x1x1]
%onnx::Conv_795[FLOAT, 184x1x5x5]
%onnx::Conv_798[FLOAT, 184x92x1x1]
%onnx::Conv_801[FLOAT, 1104x184x1x1]
%onnx::Conv_802[FLOAT, 1104]
%onnx::Conv_804[FLOAT, 1104x1x5x5]
%onnx::Conv_807[FLOAT, 184x1104x1x1]
%onnx::Conv_810[FLOAT, 184x92x1x1]
%onnx::Conv_813[FLOAT, 184x1x5x5]
%onnx::Conv_816[FLOAT, 352x92x1x1]
%onnx::Conv_817[FLOAT, 352]
%onnx::Conv_819[FLOAT, 1504x352x1x1]
%onnx::Conv_820[FLOAT, 1504]
) {
%onnx::Conv_814 = Identity(%onnx::Conv_790)
%onnx::Conv_811 = Identity(%onnx::Conv_790)
%onnx::Conv_808 = Identity(%onnx::Conv_790)
%onnx::Conv_805 = Identity(%onnx::Conv_802)
%onnx::Conv_799 = Identity(%onnx::Conv_790)
%onnx::Conv_796 = Identity(%onnx::Conv_790)
%onnx::Conv_793 = Identity(%onnx::Conv_790)
%onnx::Conv_787 = Identity(%onnx::Conv_754)
%onnx::Conv_784 = Identity(%onnx::Conv_754)
%onnx::Conv_781 = Identity(%onnx::Conv_754)
%onnx::Conv_778 = Identity(%onnx::Conv_757)
%onnx::Conv_775 = Identity(%onnx::Conv_757)
%onnx::Conv_772 = Identity(%onnx::Conv_754)
%onnx::Conv_769 = Identity(%onnx::Conv_754)
%onnx::Conv_766 = Identity(%onnx::Conv_754)
%onnx::Conv_763 = Identity(%onnx::Conv_754)
%onnx::Conv_760 = Identity(%onnx::Conv_757)
%onnx::Conv_751 = Identity(%onnx::Conv_703)
%onnx::Conv_748 = Identity(%onnx::Conv_703)
%onnx::Conv_745 = Identity(%onnx::Conv_718)
%onnx::Conv_742 = Identity(%onnx::Conv_703)
%onnx::Conv_739 = Identity(%onnx::Conv_703)
%onnx::Conv_736 = Identity(%onnx::Conv_718)
%onnx::Conv_733 = Identity(%onnx::Conv_718)
%onnx::Conv_730 = Identity(%onnx::Conv_718)
%onnx::Conv_727 = Identity(%onnx::Conv_718)
%onnx::Conv_724 = Identity(%onnx::Conv_718)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%onnx::Conv_715 = Identity(%onnx::Conv_712)
%onnx::Conv_709 = Identity(%onnx::Conv_691)
%onnx::Conv_706 = Identity(%onnx::Conv_703)
%onnx::Conv_700 = Identity(%onnx::Conv_691)
%onnx::Conv_697 = Identity(%onnx::Conv_691)
%onnx::Conv_694 = Identity(%onnx::Conv_691)
%onnx::Conv_688 = Identity(%onnx::Conv_661)
%onnx::Conv_685 = Identity(%onnx::Conv_661)
%onnx::Conv_682 = Identity(%onnx::Conv_661)
%onnx::Conv_679 = Identity(%onnx::Conv_661)
%onnx::Conv_676 = Identity(%onnx::Conv_661)
%onnx::Conv_673 = Identity(%onnx::Conv_661)
%onnx::Conv_670 = Identity(%onnx::Conv_661)
%onnx::Conv_667 = Identity(%onnx::Conv_664)
%onnx::Conv_658 = Identity(%onnx::Conv_643)
%onnx::Conv_655 = Identity(%onnx::Conv_643)
%onnx::Conv_652 = Identity(%onnx::Conv_643)
%onnx::Conv_649 = Identity(%onnx::Conv_643)
%onnx::Conv_646 = Identity(%onnx::Conv_643)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_642, %onnx::Conv_643)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.4/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%640 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %640
}
|
val_accuracy
| 0 | 68,051,584 | 1,720,388 |
{'zcp_synflow': 73.72830480008375, 'zcp_zen': 64.05784606933594, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.0987391248345375, 'zcp_flops': 68051584.0, 'zcp_grad_norm': 20.486936569213867, 'zcp_grasp': 0.07383918762207031, 'zcp_jacov': -16.050730575181774, 'zcp_l2_norm': 573.656005859375, 'zcp_nwot': 210.6795147563435, 'zcp_params': 1720388.0, 'zcp_plain': 0.0012452179798856378, 'zcp_snip': 36.01094055175781, 'lat_1080ti_1': 0.5536558861345441, 'lat_1080ti_32': 0.46264975127909597, 'lat_1080ti_64': 0.42037552251697097, 'lat_2080ti_1': 0.5629696369602291, 'lat_2080ti_32': 0.4751209458609458, 'lat_2080ti_64': 0.4165617183383229, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.3889021326536225, 'lat_fpga': 0.49496828585281866, 'lat_gold_6226': 0.32085138352514914, 'lat_gold_6240': 0.42562896178607806, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.40303774908928625, 'lat_raspi4': 0.4388291995950836, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.5433070866141733, 'lat_silver_4114': 0.483392320760981, 'lat_silver_4210r': 0.4956546483371446, 'lat_titan_rtx_1': 0.5263872715241379, 'lat_titan_rtx_32': 0.47116232738518915, 'lat_titan_rtx_64': 0.4330023100035318, 'lat_titanx_1': 0.2814742618288809, 'lat_titanx_32': 0.4686185199570219, 'lat_titanx_64': 0.40585234648829155, 'lat_titanxp_1': 0.5071228398138783, 'lat_titanxp_32': 0.45308177822770007, 'lat_titanxp_64': 0.4197119371106256}
| |
FBNet_698
|
FBNet
|
698
|
698
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_723[FLOAT, 16x3x3x3]
%onnx::Conv_724[FLOAT, 16]
%onnx::Conv_726[FLOAT, 96x16x1x1]
%onnx::Conv_727[FLOAT, 96]
%onnx::Conv_729[FLOAT, 96x1x3x3]
%onnx::Conv_732[FLOAT, 16x96x1x1]
%onnx::Conv_735[FLOAT, 48x16x1x1]
%onnx::Conv_736[FLOAT, 48]
%onnx::Conv_738[FLOAT, 48x1x3x3]
%onnx::Conv_741[FLOAT, 24x48x1x1]
%onnx::Conv_742[FLOAT, 24]
%onnx::Conv_744[FLOAT, 24x12x1x1]
%onnx::Conv_747[FLOAT, 24x1x5x5]
%onnx::Conv_750[FLOAT, 24x12x1x1]
%onnx::Conv_753[FLOAT, 72x24x1x1]
%onnx::Conv_754[FLOAT, 72]
%onnx::Conv_756[FLOAT, 72x1x5x5]
%onnx::Conv_759[FLOAT, 24x72x1x1]
%onnx::Conv_762[FLOAT, 144x24x1x1]
%onnx::Conv_763[FLOAT, 144]
%onnx::Conv_765[FLOAT, 144x1x5x5]
%onnx::Conv_768[FLOAT, 24x144x1x1]
%onnx::Conv_771[FLOAT, 24x12x1x1]
%onnx::Conv_774[FLOAT, 24x1x3x3]
%onnx::Conv_777[FLOAT, 32x12x1x1]
%onnx::Conv_778[FLOAT, 32]
%onnx::Conv_780[FLOAT, 32x32x1x1]
%onnx::Conv_783[FLOAT, 32x1x3x3]
%onnx::Conv_786[FLOAT, 32x32x1x1]
%onnx::Conv_789[FLOAT, 32x16x1x1]
%onnx::Conv_792[FLOAT, 32x1x5x5]
%onnx::Conv_795[FLOAT, 32x16x1x1]
%onnx::Conv_798[FLOAT, 192x32x1x1]
%onnx::Conv_799[FLOAT, 192]
%onnx::Conv_801[FLOAT, 192x1x5x5]
%onnx::Conv_804[FLOAT, 64x192x1x1]
%onnx::Conv_805[FLOAT, 64]
%onnx::Conv_807[FLOAT, 64x32x1x1]
%onnx::Conv_810[FLOAT, 64x1x3x3]
%onnx::Conv_813[FLOAT, 64x32x1x1]
%onnx::Conv_816[FLOAT, 64x64x1x1]
%onnx::Conv_819[FLOAT, 64x1x5x5]
%onnx::Conv_822[FLOAT, 64x64x1x1]
%onnx::Conv_825[FLOAT, 64x64x1x1]
%onnx::Conv_828[FLOAT, 64x1x5x5]
%onnx::Conv_831[FLOAT, 64x64x1x1]
%onnx::Conv_834[FLOAT, 64x32x1x1]
%onnx::Conv_837[FLOAT, 64x1x3x3]
%onnx::Conv_840[FLOAT, 112x32x1x1]
%onnx::Conv_841[FLOAT, 112]
%onnx::Conv_843[FLOAT, 336x112x1x1]
%onnx::Conv_844[FLOAT, 336]
%onnx::Conv_846[FLOAT, 336x1x5x5]
%onnx::Conv_849[FLOAT, 112x336x1x1]
%onnx::Conv_852[FLOAT, 112x56x1x1]
%onnx::Conv_855[FLOAT, 112x1x3x3]
%onnx::Conv_858[FLOAT, 112x56x1x1]
%onnx::Conv_861[FLOAT, 336x112x1x1]
%onnx::Conv_864[FLOAT, 336x1x3x3]
%onnx::Conv_867[FLOAT, 112x336x1x1]
%onnx::Conv_870[FLOAT, 672x112x1x1]
%onnx::Conv_871[FLOAT, 672]
%onnx::Conv_873[FLOAT, 672x1x5x5]
%onnx::Conv_876[FLOAT, 184x672x1x1]
%onnx::Conv_877[FLOAT, 184]
%onnx::Conv_879[FLOAT, 184x184x1x1]
%onnx::Conv_882[FLOAT, 184x1x3x3]
%onnx::Conv_885[FLOAT, 184x184x1x1]
%onnx::Conv_888[FLOAT, 1104x184x1x1]
%onnx::Conv_889[FLOAT, 1104]
%onnx::Conv_891[FLOAT, 1104x1x3x3]
%onnx::Conv_894[FLOAT, 184x1104x1x1]
%onnx::Conv_897[FLOAT, 184x92x1x1]
%onnx::Conv_900[FLOAT, 184x1x5x5]
%onnx::Conv_903[FLOAT, 184x92x1x1]
%onnx::Conv_906[FLOAT, 184x184x1x1]
%onnx::Conv_909[FLOAT, 184x1x5x5]
%onnx::Conv_912[FLOAT, 352x184x1x1]
%onnx::Conv_913[FLOAT, 352]
%onnx::Conv_915[FLOAT, 1504x352x1x1]
%onnx::Conv_916[FLOAT, 1504]
) {
%onnx::Conv_910 = Identity(%onnx::Conv_877)
%onnx::Conv_907 = Identity(%onnx::Conv_877)
%onnx::Conv_904 = Identity(%onnx::Conv_877)
%onnx::Conv_901 = Identity(%onnx::Conv_877)
%onnx::Conv_898 = Identity(%onnx::Conv_877)
%onnx::Conv_895 = Identity(%onnx::Conv_877)
%onnx::Conv_892 = Identity(%onnx::Conv_889)
%onnx::Conv_886 = Identity(%onnx::Conv_877)
%onnx::Conv_883 = Identity(%onnx::Conv_877)
%onnx::Conv_880 = Identity(%onnx::Conv_877)
%onnx::Conv_874 = Identity(%onnx::Conv_871)
%onnx::Conv_868 = Identity(%onnx::Conv_841)
%onnx::Conv_865 = Identity(%onnx::Conv_844)
%onnx::Conv_862 = Identity(%onnx::Conv_844)
%onnx::Conv_859 = Identity(%onnx::Conv_841)
%onnx::Conv_856 = Identity(%onnx::Conv_841)
%onnx::Conv_853 = Identity(%onnx::Conv_841)
%onnx::Conv_850 = Identity(%onnx::Conv_841)
%onnx::Conv_847 = Identity(%onnx::Conv_844)
%onnx::Conv_838 = Identity(%onnx::Conv_805)
%onnx::Conv_835 = Identity(%onnx::Conv_805)
%onnx::Conv_832 = Identity(%onnx::Conv_805)
%onnx::Conv_829 = Identity(%onnx::Conv_805)
%onnx::Conv_826 = Identity(%onnx::Conv_805)
%onnx::Conv_823 = Identity(%onnx::Conv_805)
%onnx::Conv_820 = Identity(%onnx::Conv_805)
%onnx::Conv_817 = Identity(%onnx::Conv_805)
%onnx::Conv_814 = Identity(%onnx::Conv_805)
%onnx::Conv_811 = Identity(%onnx::Conv_805)
%onnx::Conv_808 = Identity(%onnx::Conv_805)
%onnx::Conv_802 = Identity(%onnx::Conv_799)
%onnx::Conv_796 = Identity(%onnx::Conv_778)
%onnx::Conv_793 = Identity(%onnx::Conv_778)
%onnx::Conv_790 = Identity(%onnx::Conv_778)
%onnx::Conv_787 = Identity(%onnx::Conv_778)
%onnx::Conv_784 = Identity(%onnx::Conv_778)
%onnx::Conv_781 = Identity(%onnx::Conv_778)
%onnx::Conv_775 = Identity(%onnx::Conv_742)
%onnx::Conv_772 = Identity(%onnx::Conv_742)
%onnx::Conv_769 = Identity(%onnx::Conv_742)
%onnx::Conv_766 = Identity(%onnx::Conv_763)
%onnx::Conv_760 = Identity(%onnx::Conv_742)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_751 = Identity(%onnx::Conv_742)
%onnx::Conv_748 = Identity(%onnx::Conv_742)
%onnx::Conv_745 = Identity(%onnx::Conv_742)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_733 = Identity(%onnx::Conv_724)
%onnx::Conv_730 = Identity(%onnx::Conv_727)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_723, %onnx::Conv_724)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_882, %onnx::Conv_883)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_885, %onnx::Conv_886)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_888, %onnx::Conv_889)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_891, %onnx::Conv_892)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_894, %onnx::Conv_895)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_897, %onnx::Conv_898)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_900, %onnx::Conv_901)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_903, %onnx::Conv_904)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_906, %onnx::Conv_907)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_909, %onnx::Conv_910)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_912, %onnx::Conv_913)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_915, %onnx::Conv_916)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%721 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %721
}
|
val_accuracy
| 0 | 70,698,112 | 1,803,956 |
{'zcp_synflow': 76.83113673092596, 'zcp_zen': 68.79041290283203, 'zcp_epe_nas': 6.18065685647159, 'zcp_fisher': 0.22941865026950836, 'zcp_flops': 70698112.0, 'zcp_grad_norm': 29.293079376220703, 'zcp_grasp': -0.30115509033203125, 'zcp_jacov': -16.067297550702264, 'zcp_l2_norm': 613.7181396484375, 'zcp_nwot': 214.34502642705868, 'zcp_params': 1803956.0, 'zcp_plain': -0.0015051525551825762, 'zcp_snip': 47.72911834716797, 'lat_1080ti_1': 0.661558596672693, 'lat_1080ti_32': 0.7234863347693629, 'lat_1080ti_64': 0.6275404615596263, 'lat_2080ti_1': 0.7496229353482465, 'lat_2080ti_32': 0.7628208770435523, 'lat_2080ti_64': 0.6371215549623563, 'lat_essential_ph_1': 0.4528301886792453, 'lat_eyeriss': 0.4932976362936206, 'lat_fpga': 0.4416331054136691, 'lat_gold_6226': 0.29970734101443264, 'lat_gold_6240': 0.5336892952343891, 'lat_pixel2': 0.5, 'lat_pixel3': 0.5271449504382059, 'lat_raspi4': 0.5161437104917298, 'lat_samsung_a50': 0.22105263157894736, 'lat_samsung_s7': 0.23622047244094488, 'lat_silver_4114': 0.5915773894140242, 'lat_silver_4210r': 0.6108721858680406, 'lat_titan_rtx_1': 0.7262829369337233, 'lat_titan_rtx_32': 0.7485248990918452, 'lat_titan_rtx_64': 0.6901216886529947, 'lat_titanx_1': 0.40186216031230376, 'lat_titanx_32': 0.7133622981793484, 'lat_titanx_64': 0.6068427079945082, 'lat_titanxp_1': 0.6819981043944334, 'lat_titanxp_32': 0.7319123009068024, 'lat_titanxp_64': 0.6747900109541934}
| |
FBNet_2679
|
FBNet
|
2679
|
2679
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_642[FLOAT, 16x3x3x3]
%onnx::Conv_643[FLOAT, 16]
%onnx::Conv_645[FLOAT, 16x16x1x1]
%onnx::Conv_648[FLOAT, 16x1x3x3]
%onnx::Conv_651[FLOAT, 16x16x1x1]
%onnx::Conv_654[FLOAT, 96x16x1x1]
%onnx::Conv_655[FLOAT, 96]
%onnx::Conv_657[FLOAT, 96x1x3x3]
%onnx::Conv_660[FLOAT, 24x96x1x1]
%onnx::Conv_661[FLOAT, 24]
%onnx::Conv_663[FLOAT, 72x24x1x1]
%onnx::Conv_664[FLOAT, 72]
%onnx::Conv_666[FLOAT, 72x1x5x5]
%onnx::Conv_669[FLOAT, 24x72x1x1]
%onnx::Conv_672[FLOAT, 72x24x1x1]
%onnx::Conv_675[FLOAT, 72x1x3x3]
%onnx::Conv_678[FLOAT, 24x72x1x1]
%onnx::Conv_681[FLOAT, 72x24x1x1]
%onnx::Conv_684[FLOAT, 72x1x3x3]
%onnx::Conv_687[FLOAT, 24x72x1x1]
%onnx::Conv_690[FLOAT, 24x12x1x1]
%onnx::Conv_693[FLOAT, 24x1x5x5]
%onnx::Conv_696[FLOAT, 32x12x1x1]
%onnx::Conv_697[FLOAT, 32]
%onnx::Conv_699[FLOAT, 32x16x1x1]
%onnx::Conv_702[FLOAT, 32x1x3x3]
%onnx::Conv_705[FLOAT, 32x16x1x1]
%onnx::Conv_708[FLOAT, 32x16x1x1]
%onnx::Conv_711[FLOAT, 32x1x5x5]
%onnx::Conv_714[FLOAT, 32x16x1x1]
%onnx::Conv_717[FLOAT, 32x32x1x1]
%onnx::Conv_720[FLOAT, 32x1x3x3]
%onnx::Conv_723[FLOAT, 64x32x1x1]
%onnx::Conv_724[FLOAT, 64]
%onnx::Conv_726[FLOAT, 64x64x1x1]
%onnx::Conv_729[FLOAT, 64x1x3x3]
%onnx::Conv_732[FLOAT, 64x64x1x1]
%onnx::Conv_735[FLOAT, 192x64x1x1]
%onnx::Conv_736[FLOAT, 192]
%onnx::Conv_738[FLOAT, 192x1x3x3]
%onnx::Conv_741[FLOAT, 64x192x1x1]
%onnx::Conv_744[FLOAT, 192x64x1x1]
%onnx::Conv_747[FLOAT, 192x1x5x5]
%onnx::Conv_750[FLOAT, 64x192x1x1]
%onnx::Conv_753[FLOAT, 64x64x1x1]
%onnx::Conv_756[FLOAT, 64x1x5x5]
%onnx::Conv_759[FLOAT, 112x64x1x1]
%onnx::Conv_760[FLOAT, 112]
%onnx::Conv_762[FLOAT, 336x112x1x1]
%onnx::Conv_763[FLOAT, 336]
%onnx::Conv_765[FLOAT, 336x1x5x5]
%onnx::Conv_768[FLOAT, 112x336x1x1]
%onnx::Conv_771[FLOAT, 112x112x1x1]
%onnx::Conv_774[FLOAT, 112x1x5x5]
%onnx::Conv_777[FLOAT, 112x112x1x1]
%onnx::Conv_780[FLOAT, 112x56x1x1]
%onnx::Conv_783[FLOAT, 112x1x5x5]
%onnx::Conv_786[FLOAT, 112x56x1x1]
%onnx::Conv_789[FLOAT, 184x112x1x1]
%onnx::Conv_790[FLOAT, 184]
%onnx::Conv_792[FLOAT, 184x92x1x1]
%onnx::Conv_795[FLOAT, 184x1x3x3]
%onnx::Conv_798[FLOAT, 184x92x1x1]
%onnx::Conv_801[FLOAT, 184x184x1x1]
%onnx::Conv_804[FLOAT, 184x1x3x3]
%onnx::Conv_807[FLOAT, 184x184x1x1]
%onnx::Conv_810[FLOAT, 1104x184x1x1]
%onnx::Conv_811[FLOAT, 1104]
%onnx::Conv_813[FLOAT, 1104x1x3x3]
%onnx::Conv_816[FLOAT, 352x1104x1x1]
%onnx::Conv_817[FLOAT, 352]
%onnx::Conv_819[FLOAT, 1504x352x1x1]
%onnx::Conv_820[FLOAT, 1504]
) {
%onnx::Conv_814 = Identity(%onnx::Conv_811)
%onnx::Conv_808 = Identity(%onnx::Conv_790)
%onnx::Conv_805 = Identity(%onnx::Conv_790)
%onnx::Conv_802 = Identity(%onnx::Conv_790)
%onnx::Conv_799 = Identity(%onnx::Conv_790)
%onnx::Conv_796 = Identity(%onnx::Conv_790)
%onnx::Conv_793 = Identity(%onnx::Conv_790)
%onnx::Conv_787 = Identity(%onnx::Conv_760)
%onnx::Conv_784 = Identity(%onnx::Conv_760)
%onnx::Conv_781 = Identity(%onnx::Conv_760)
%onnx::Conv_778 = Identity(%onnx::Conv_760)
%onnx::Conv_775 = Identity(%onnx::Conv_760)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_763)
%onnx::Conv_757 = Identity(%onnx::Conv_724)
%onnx::Conv_754 = Identity(%onnx::Conv_724)
%onnx::Conv_751 = Identity(%onnx::Conv_724)
%onnx::Conv_748 = Identity(%onnx::Conv_736)
%onnx::Conv_745 = Identity(%onnx::Conv_736)
%onnx::Conv_742 = Identity(%onnx::Conv_724)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_733 = Identity(%onnx::Conv_724)
%onnx::Conv_730 = Identity(%onnx::Conv_724)
%onnx::Conv_727 = Identity(%onnx::Conv_724)
%onnx::Conv_721 = Identity(%onnx::Conv_697)
%onnx::Conv_718 = Identity(%onnx::Conv_697)
%onnx::Conv_715 = Identity(%onnx::Conv_697)
%onnx::Conv_712 = Identity(%onnx::Conv_697)
%onnx::Conv_709 = Identity(%onnx::Conv_697)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_697)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%onnx::Conv_694 = Identity(%onnx::Conv_661)
%onnx::Conv_691 = Identity(%onnx::Conv_661)
%onnx::Conv_688 = Identity(%onnx::Conv_661)
%onnx::Conv_685 = Identity(%onnx::Conv_664)
%onnx::Conv_682 = Identity(%onnx::Conv_664)
%onnx::Conv_679 = Identity(%onnx::Conv_661)
%onnx::Conv_676 = Identity(%onnx::Conv_664)
%onnx::Conv_673 = Identity(%onnx::Conv_664)
%onnx::Conv_670 = Identity(%onnx::Conv_661)
%onnx::Conv_667 = Identity(%onnx::Conv_664)
%onnx::Conv_658 = Identity(%onnx::Conv_655)
%onnx::Conv_652 = Identity(%onnx::Conv_643)
%onnx::Conv_649 = Identity(%onnx::Conv_643)
%onnx::Conv_646 = Identity(%onnx::Conv_643)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_642, %onnx::Conv_643)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%640 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %640
}
|
val_accuracy
| 0 | 57,662,464 | 1,656,948 |
{'zcp_synflow': 74.05001067786424, 'zcp_zen': 61.62700653076172, 'zcp_epe_nas': 15.41013582781605, 'zcp_fisher': 0.11820351332426071, 'zcp_flops': 57662464.0, 'zcp_grad_norm': 21.754865646362305, 'zcp_grasp': 0.10499191284179688, 'zcp_jacov': -16.05589396036223, 'zcp_l2_norm': 537.7435913085938, 'zcp_nwot': 210.95617043825578, 'zcp_params': 1656948.0, 'zcp_plain': -0.0012475901748985052, 'zcp_snip': 35.74563980102539, 'lat_1080ti_1': 0.5882727282864206, 'lat_1080ti_32': 0.4862368246002621, 'lat_1080ti_64': 0.4093052812413182, 'lat_2080ti_1': 0.5560288968274725, 'lat_2080ti_32': 0.49832576897407904, 'lat_2080ti_64': 0.459476092902443, 'lat_essential_ph_1': 0.1509433962264151, 'lat_eyeriss': 0.3052280270661801, 'lat_fpga': 0.3564650352764225, 'lat_gold_6226': 0.1846512338024255, 'lat_gold_6240': 0.335012611795281, 'lat_pixel2': 0.17391304347826086, 'lat_pixel3': 0.3199375040881169, 'lat_raspi4': 0.36342584948542656, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.11023622047244094, 'lat_silver_4114': 0.3469343181367638, 'lat_silver_4210r': 0.39931498579505176, 'lat_titan_rtx_1': 0.5368843603085829, 'lat_titan_rtx_32': 0.4737666810202344, 'lat_titan_rtx_64': 0.46725870906565903, 'lat_titanx_1': 0.28683955660891686, 'lat_titanx_32': 0.46282950304183934, 'lat_titanx_64': 0.39085329082717973, 'lat_titanxp_1': 0.4929371955420349, 'lat_titanxp_32': 0.4752314319474367, 'lat_titanxp_64': 0.43044168245377806}
| |
FBNet_4021
|
FBNet
|
4021
|
4021
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_612[FLOAT, 16x3x3x3]
%onnx::Conv_613[FLOAT, 16]
%onnx::Conv_615[FLOAT, 48x16x1x1]
%onnx::Conv_616[FLOAT, 48]
%onnx::Conv_618[FLOAT, 48x1x5x5]
%onnx::Conv_621[FLOAT, 16x48x1x1]
%onnx::Conv_624[FLOAT, 16x16x1x1]
%onnx::Conv_627[FLOAT, 16x1x3x3]
%onnx::Conv_630[FLOAT, 24x16x1x1]
%onnx::Conv_631[FLOAT, 24]
%onnx::Conv_633[FLOAT, 144x24x1x1]
%onnx::Conv_634[FLOAT, 144]
%onnx::Conv_636[FLOAT, 144x1x3x3]
%onnx::Conv_639[FLOAT, 24x144x1x1]
%onnx::Conv_642[FLOAT, 24x24x1x1]
%onnx::Conv_645[FLOAT, 24x1x5x5]
%onnx::Conv_648[FLOAT, 24x24x1x1]
%onnx::Conv_651[FLOAT, 144x24x1x1]
%onnx::Conv_654[FLOAT, 144x1x5x5]
%onnx::Conv_657[FLOAT, 32x144x1x1]
%onnx::Conv_658[FLOAT, 32]
%onnx::Conv_660[FLOAT, 32x16x1x1]
%onnx::Conv_663[FLOAT, 32x1x5x5]
%onnx::Conv_666[FLOAT, 32x16x1x1]
%onnx::Conv_669[FLOAT, 32x16x1x1]
%onnx::Conv_672[FLOAT, 32x1x5x5]
%onnx::Conv_675[FLOAT, 32x16x1x1]
%onnx::Conv_678[FLOAT, 96x32x1x1]
%onnx::Conv_679[FLOAT, 96]
%onnx::Conv_681[FLOAT, 96x1x5x5]
%onnx::Conv_684[FLOAT, 32x96x1x1]
%onnx::Conv_687[FLOAT, 32x32x1x1]
%onnx::Conv_690[FLOAT, 32x1x3x3]
%onnx::Conv_693[FLOAT, 64x32x1x1]
%onnx::Conv_694[FLOAT, 64]
%onnx::Conv_696[FLOAT, 384x64x1x1]
%onnx::Conv_697[FLOAT, 384]
%onnx::Conv_699[FLOAT, 384x1x5x5]
%onnx::Conv_702[FLOAT, 64x384x1x1]
%onnx::Conv_705[FLOAT, 192x64x1x1]
%onnx::Conv_706[FLOAT, 192]
%onnx::Conv_708[FLOAT, 192x1x3x3]
%onnx::Conv_711[FLOAT, 64x192x1x1]
%onnx::Conv_714[FLOAT, 384x64x1x1]
%onnx::Conv_717[FLOAT, 384x1x5x5]
%onnx::Conv_720[FLOAT, 112x384x1x1]
%onnx::Conv_721[FLOAT, 112]
%onnx::Conv_723[FLOAT, 112x112x1x1]
%onnx::Conv_726[FLOAT, 112x1x5x5]
%onnx::Conv_729[FLOAT, 112x112x1x1]
%onnx::Conv_732[FLOAT, 112x56x1x1]
%onnx::Conv_735[FLOAT, 112x1x3x3]
%onnx::Conv_738[FLOAT, 112x56x1x1]
%onnx::Conv_741[FLOAT, 112x56x1x1]
%onnx::Conv_744[FLOAT, 112x1x5x5]
%onnx::Conv_747[FLOAT, 112x56x1x1]
%onnx::Conv_750[FLOAT, 336x112x1x1]
%onnx::Conv_751[FLOAT, 336]
%onnx::Conv_753[FLOAT, 336x1x3x3]
%onnx::Conv_756[FLOAT, 184x336x1x1]
%onnx::Conv_757[FLOAT, 184]
%onnx::Conv_759[FLOAT, 1104x184x1x1]
%onnx::Conv_760[FLOAT, 1104]
%onnx::Conv_762[FLOAT, 1104x1x5x5]
%onnx::Conv_765[FLOAT, 184x1104x1x1]
%onnx::Conv_768[FLOAT, 1104x184x1x1]
%onnx::Conv_771[FLOAT, 1104x1x5x5]
%onnx::Conv_774[FLOAT, 184x1104x1x1]
%onnx::Conv_777[FLOAT, 184x184x1x1]
%onnx::Conv_780[FLOAT, 184x1x5x5]
%onnx::Conv_783[FLOAT, 352x184x1x1]
%onnx::Conv_784[FLOAT, 352]
%onnx::Conv_786[FLOAT, 1504x352x1x1]
%onnx::Conv_787[FLOAT, 1504]
) {
%onnx::Conv_781 = Identity(%onnx::Conv_757)
%onnx::Conv_778 = Identity(%onnx::Conv_757)
%onnx::Conv_775 = Identity(%onnx::Conv_757)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_757)
%onnx::Conv_763 = Identity(%onnx::Conv_760)
%onnx::Conv_754 = Identity(%onnx::Conv_751)
%onnx::Conv_748 = Identity(%onnx::Conv_721)
%onnx::Conv_745 = Identity(%onnx::Conv_721)
%onnx::Conv_742 = Identity(%onnx::Conv_721)
%onnx::Conv_739 = Identity(%onnx::Conv_721)
%onnx::Conv_736 = Identity(%onnx::Conv_721)
%onnx::Conv_733 = Identity(%onnx::Conv_721)
%onnx::Conv_730 = Identity(%onnx::Conv_721)
%onnx::Conv_727 = Identity(%onnx::Conv_721)
%onnx::Conv_724 = Identity(%onnx::Conv_721)
%onnx::Conv_718 = Identity(%onnx::Conv_697)
%onnx::Conv_715 = Identity(%onnx::Conv_697)
%onnx::Conv_712 = Identity(%onnx::Conv_694)
%onnx::Conv_709 = Identity(%onnx::Conv_706)
%onnx::Conv_703 = Identity(%onnx::Conv_694)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%onnx::Conv_691 = Identity(%onnx::Conv_658)
%onnx::Conv_688 = Identity(%onnx::Conv_658)
%onnx::Conv_685 = Identity(%onnx::Conv_658)
%onnx::Conv_682 = Identity(%onnx::Conv_679)
%onnx::Conv_676 = Identity(%onnx::Conv_658)
%onnx::Conv_673 = Identity(%onnx::Conv_658)
%onnx::Conv_670 = Identity(%onnx::Conv_658)
%onnx::Conv_667 = Identity(%onnx::Conv_658)
%onnx::Conv_664 = Identity(%onnx::Conv_658)
%onnx::Conv_661 = Identity(%onnx::Conv_658)
%onnx::Conv_655 = Identity(%onnx::Conv_634)
%onnx::Conv_652 = Identity(%onnx::Conv_634)
%onnx::Conv_649 = Identity(%onnx::Conv_631)
%onnx::Conv_646 = Identity(%onnx::Conv_631)
%onnx::Conv_643 = Identity(%onnx::Conv_631)
%onnx::Conv_640 = Identity(%onnx::Conv_631)
%onnx::Conv_637 = Identity(%onnx::Conv_634)
%onnx::Conv_628 = Identity(%onnx::Conv_613)
%onnx::Conv_625 = Identity(%onnx::Conv_613)
%onnx::Conv_622 = Identity(%onnx::Conv_613)
%onnx::Conv_619 = Identity(%onnx::Conv_616)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_612, %onnx::Conv_613)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_615, %onnx::Conv_616)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%610 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %610
}
|
val_accuracy
| 0 | 68,026,752 | 2,038,436 |
{'zcp_synflow': 74.12439005504446, 'zcp_zen': 65.19499969482422, 'zcp_epe_nas': 13.949467158276398, 'zcp_fisher': 0.09350922703742981, 'zcp_flops': 68026752.0, 'zcp_grad_norm': 24.560834884643555, 'zcp_grasp': -0.04692268371582031, 'zcp_jacov': -16.05795595518791, 'zcp_l2_norm': 597.448974609375, 'zcp_nwot': 212.3915334213015, 'zcp_params': 2038436.0, 'zcp_plain': -0.00321654649451375, 'zcp_snip': 40.53134536743164, 'lat_1080ti_1': 0.5631961359631636, 'lat_1080ti_32': 0.45802772867179364, 'lat_1080ti_64': 0.43198935388886767, 'lat_2080ti_1': 0.5103914370007541, 'lat_2080ti_32': 0.4893199741650746, 'lat_2080ti_64': 0.427388401238235, 'lat_essential_ph_1': 0.22641509433962265, 'lat_eyeriss': 0.46510369867786455, 'lat_fpga': 0.43354871906510034, 'lat_gold_6226': 0.40062244911250705, 'lat_gold_6240': 0.44721232919447657, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.4457135251132335, 'lat_raspi4': 0.4914157892544112, 'lat_samsung_a50': 0.21052631578947367, 'lat_samsung_s7': 0.11811023622047244, 'lat_silver_4114': 0.45925033969514184, 'lat_silver_4210r': 0.47900564831820913, 'lat_titan_rtx_1': 0.47600372696952115, 'lat_titan_rtx_32': 0.46132893622572413, 'lat_titan_rtx_64': 0.42670471273322386, 'lat_titanx_1': 0.24690679434503088, 'lat_titanx_32': 0.44854566159340264, 'lat_titanx_64': 0.410093557843048, 'lat_titanxp_1': 0.4532232362017128, 'lat_titanxp_32': 0.45400635391943595, 'lat_titanxp_64': 0.42700830708028525}
| |
FBNet_386
|
FBNet
|
386
|
386
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_605[FLOAT, 16x3x3x3]
%onnx::Conv_606[FLOAT, 16]
%onnx::Conv_608[FLOAT, 16x16x1x1]
%onnx::Conv_611[FLOAT, 16x1x5x5]
%onnx::Conv_614[FLOAT, 16x16x1x1]
%onnx::Conv_617[FLOAT, 48x16x1x1]
%onnx::Conv_618[FLOAT, 48]
%onnx::Conv_620[FLOAT, 48x1x3x3]
%onnx::Conv_623[FLOAT, 24x48x1x1]
%onnx::Conv_624[FLOAT, 24]
%onnx::Conv_626[FLOAT, 72x24x1x1]
%onnx::Conv_627[FLOAT, 72]
%onnx::Conv_629[FLOAT, 72x1x5x5]
%onnx::Conv_632[FLOAT, 24x72x1x1]
%onnx::Conv_635[FLOAT, 72x24x1x1]
%onnx::Conv_638[FLOAT, 72x1x3x3]
%onnx::Conv_641[FLOAT, 24x72x1x1]
%onnx::Conv_644[FLOAT, 144x24x1x1]
%onnx::Conv_645[FLOAT, 144]
%onnx::Conv_647[FLOAT, 144x1x5x5]
%onnx::Conv_650[FLOAT, 24x144x1x1]
%onnx::Conv_653[FLOAT, 72x24x1x1]
%onnx::Conv_656[FLOAT, 72x1x3x3]
%onnx::Conv_659[FLOAT, 32x72x1x1]
%onnx::Conv_660[FLOAT, 32]
%onnx::Conv_662[FLOAT, 192x32x1x1]
%onnx::Conv_663[FLOAT, 192]
%onnx::Conv_665[FLOAT, 192x1x3x3]
%onnx::Conv_668[FLOAT, 32x192x1x1]
%onnx::Conv_671[FLOAT, 32x32x1x1]
%onnx::Conv_674[FLOAT, 32x1x3x3]
%onnx::Conv_677[FLOAT, 32x32x1x1]
%onnx::Conv_680[FLOAT, 96x32x1x1]
%onnx::Conv_681[FLOAT, 96]
%onnx::Conv_683[FLOAT, 96x1x5x5]
%onnx::Conv_686[FLOAT, 64x96x1x1]
%onnx::Conv_687[FLOAT, 64]
%onnx::Conv_689[FLOAT, 64x32x1x1]
%onnx::Conv_692[FLOAT, 64x1x3x3]
%onnx::Conv_695[FLOAT, 64x32x1x1]
%onnx::Conv_698[FLOAT, 64x32x1x1]
%onnx::Conv_701[FLOAT, 64x1x3x3]
%onnx::Conv_704[FLOAT, 64x32x1x1]
%onnx::Conv_707[FLOAT, 112x64x1x1]
%onnx::Conv_708[FLOAT, 112]
%onnx::Conv_710[FLOAT, 336x112x1x1]
%onnx::Conv_711[FLOAT, 336]
%onnx::Conv_713[FLOAT, 336x1x5x5]
%onnx::Conv_716[FLOAT, 112x336x1x1]
%onnx::Conv_719[FLOAT, 672x112x1x1]
%onnx::Conv_720[FLOAT, 672]
%onnx::Conv_722[FLOAT, 672x1x5x5]
%onnx::Conv_725[FLOAT, 112x672x1x1]
%onnx::Conv_728[FLOAT, 112x112x1x1]
%onnx::Conv_731[FLOAT, 112x1x3x3]
%onnx::Conv_734[FLOAT, 112x112x1x1]
%onnx::Conv_737[FLOAT, 112x112x1x1]
%onnx::Conv_740[FLOAT, 112x1x3x3]
%onnx::Conv_743[FLOAT, 184x112x1x1]
%onnx::Conv_744[FLOAT, 184]
%onnx::Conv_746[FLOAT, 552x184x1x1]
%onnx::Conv_747[FLOAT, 552]
%onnx::Conv_749[FLOAT, 552x1x3x3]
%onnx::Conv_752[FLOAT, 184x552x1x1]
%onnx::Conv_755[FLOAT, 184x92x1x1]
%onnx::Conv_758[FLOAT, 184x1x5x5]
%onnx::Conv_761[FLOAT, 184x92x1x1]
%onnx::Conv_764[FLOAT, 184x92x1x1]
%onnx::Conv_767[FLOAT, 184x1x5x5]
%onnx::Conv_770[FLOAT, 184x92x1x1]
%onnx::Conv_773[FLOAT, 352x184x1x1]
%onnx::Conv_774[FLOAT, 352]
%onnx::Conv_776[FLOAT, 1504x352x1x1]
%onnx::Conv_777[FLOAT, 1504]
) {
%onnx::Conv_771 = Identity(%onnx::Conv_744)
%onnx::Conv_768 = Identity(%onnx::Conv_744)
%onnx::Conv_765 = Identity(%onnx::Conv_744)
%onnx::Conv_762 = Identity(%onnx::Conv_744)
%onnx::Conv_759 = Identity(%onnx::Conv_744)
%onnx::Conv_756 = Identity(%onnx::Conv_744)
%onnx::Conv_753 = Identity(%onnx::Conv_744)
%onnx::Conv_750 = Identity(%onnx::Conv_747)
%onnx::Conv_741 = Identity(%onnx::Conv_708)
%onnx::Conv_738 = Identity(%onnx::Conv_708)
%onnx::Conv_735 = Identity(%onnx::Conv_708)
%onnx::Conv_732 = Identity(%onnx::Conv_708)
%onnx::Conv_729 = Identity(%onnx::Conv_708)
%onnx::Conv_726 = Identity(%onnx::Conv_708)
%onnx::Conv_723 = Identity(%onnx::Conv_720)
%onnx::Conv_717 = Identity(%onnx::Conv_708)
%onnx::Conv_714 = Identity(%onnx::Conv_711)
%onnx::Conv_705 = Identity(%onnx::Conv_687)
%onnx::Conv_702 = Identity(%onnx::Conv_687)
%onnx::Conv_699 = Identity(%onnx::Conv_687)
%onnx::Conv_696 = Identity(%onnx::Conv_687)
%onnx::Conv_693 = Identity(%onnx::Conv_687)
%onnx::Conv_690 = Identity(%onnx::Conv_687)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_678 = Identity(%onnx::Conv_660)
%onnx::Conv_675 = Identity(%onnx::Conv_660)
%onnx::Conv_672 = Identity(%onnx::Conv_660)
%onnx::Conv_669 = Identity(%onnx::Conv_660)
%onnx::Conv_666 = Identity(%onnx::Conv_663)
%onnx::Conv_657 = Identity(%onnx::Conv_627)
%onnx::Conv_654 = Identity(%onnx::Conv_627)
%onnx::Conv_651 = Identity(%onnx::Conv_624)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%onnx::Conv_642 = Identity(%onnx::Conv_624)
%onnx::Conv_639 = Identity(%onnx::Conv_627)
%onnx::Conv_636 = Identity(%onnx::Conv_627)
%onnx::Conv_633 = Identity(%onnx::Conv_624)
%onnx::Conv_630 = Identity(%onnx::Conv_627)
%onnx::Conv_621 = Identity(%onnx::Conv_618)
%onnx::Conv_615 = Identity(%onnx::Conv_606)
%onnx::Conv_612 = Identity(%onnx::Conv_606)
%onnx::Conv_609 = Identity(%onnx::Conv_606)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_605, %onnx::Conv_606)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%603 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %603
}
|
val_accuracy
| 0 | 69,740,160 | 1,432,676 |
{'zcp_synflow': 74.09430502286037, 'zcp_zen': 62.23061752319336, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.0818728506565094, 'zcp_flops': 69740160.0, 'zcp_grad_norm': 17.316268920898438, 'zcp_grasp': -0.0069637298583984375, 'zcp_jacov': -16.0478166265396, 'zcp_l2_norm': 548.9143676757812, 'zcp_nwot': 214.7577205776002, 'zcp_params': 1432676.0, 'zcp_plain': 0.005264537874609232, 'zcp_snip': 32.85633850097656, 'lat_1080ti_1': 0.3793423014863041, 'lat_1080ti_32': 0.4655957553856726, 'lat_1080ti_64': 0.5295311436314809, 'lat_2080ti_1': 0.4747288768016367, 'lat_2080ti_32': 0.48284585180614614, 'lat_2080ti_64': 0.5296163804971188, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.43298553169640397, 'lat_fpga': 0.4289177609119425, 'lat_gold_6226': 0.21582843218686731, 'lat_gold_6240': 0.2796553017299571, 'lat_pixel2': 0.391304347826087, 'lat_pixel3': 0.47695414437992495, 'lat_raspi4': 0.42423508724997255, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.1732283464566929, 'lat_silver_4114': 0.29185955471550035, 'lat_silver_4210r': 0.2926669626503101, 'lat_titan_rtx_1': 0.442157463021797, 'lat_titan_rtx_32': 0.5495000703816296, 'lat_titan_rtx_64': 0.515149032238573, 'lat_titanx_1': 0.24997674175847498, 'lat_titanx_32': 0.4848566415420901, 'lat_titanx_64': 0.5929722757362635, 'lat_titanxp_1': 0.4344405506189402, 'lat_titanxp_32': 0.49543576241490067, 'lat_titanxp_64': 0.5467392200723565}
| |
FBNet_1994
|
FBNet
|
1994
|
1994
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_623[FLOAT, 16x3x3x3]
%onnx::Conv_624[FLOAT, 16]
%onnx::Conv_626[FLOAT, 48x16x1x1]
%onnx::Conv_627[FLOAT, 48]
%onnx::Conv_629[FLOAT, 48x1x3x3]
%onnx::Conv_632[FLOAT, 16x48x1x1]
%onnx::Conv_635[FLOAT, 24x16x1x1]
%onnx::Conv_636[FLOAT, 24]
%onnx::Conv_638[FLOAT, 24x12x1x1]
%onnx::Conv_641[FLOAT, 24x1x5x5]
%onnx::Conv_644[FLOAT, 24x12x1x1]
%onnx::Conv_647[FLOAT, 72x24x1x1]
%onnx::Conv_648[FLOAT, 72]
%onnx::Conv_650[FLOAT, 72x1x5x5]
%onnx::Conv_653[FLOAT, 24x72x1x1]
%onnx::Conv_656[FLOAT, 24x12x1x1]
%onnx::Conv_659[FLOAT, 24x1x3x3]
%onnx::Conv_662[FLOAT, 32x12x1x1]
%onnx::Conv_663[FLOAT, 32]
%onnx::Conv_665[FLOAT, 96x32x1x1]
%onnx::Conv_666[FLOAT, 96]
%onnx::Conv_668[FLOAT, 96x1x5x5]
%onnx::Conv_671[FLOAT, 32x96x1x1]
%onnx::Conv_674[FLOAT, 96x32x1x1]
%onnx::Conv_677[FLOAT, 96x1x5x5]
%onnx::Conv_680[FLOAT, 32x96x1x1]
%onnx::Conv_683[FLOAT, 32x32x1x1]
%onnx::Conv_686[FLOAT, 32x1x3x3]
%onnx::Conv_689[FLOAT, 32x32x1x1]
%onnx::Conv_692[FLOAT, 192x32x1x1]
%onnx::Conv_693[FLOAT, 192]
%onnx::Conv_695[FLOAT, 192x1x5x5]
%onnx::Conv_698[FLOAT, 64x192x1x1]
%onnx::Conv_699[FLOAT, 64]
%onnx::Conv_701[FLOAT, 384x64x1x1]
%onnx::Conv_702[FLOAT, 384]
%onnx::Conv_704[FLOAT, 384x1x3x3]
%onnx::Conv_707[FLOAT, 64x384x1x1]
%onnx::Conv_710[FLOAT, 384x64x1x1]
%onnx::Conv_713[FLOAT, 384x1x3x3]
%onnx::Conv_716[FLOAT, 64x384x1x1]
%onnx::Conv_719[FLOAT, 64x64x1x1]
%onnx::Conv_722[FLOAT, 64x1x5x5]
%onnx::Conv_725[FLOAT, 64x64x1x1]
%onnx::Conv_728[FLOAT, 384x64x1x1]
%onnx::Conv_731[FLOAT, 384x1x3x3]
%onnx::Conv_734[FLOAT, 112x384x1x1]
%onnx::Conv_735[FLOAT, 112]
%onnx::Conv_737[FLOAT, 112x56x1x1]
%onnx::Conv_740[FLOAT, 112x1x3x3]
%onnx::Conv_743[FLOAT, 112x56x1x1]
%onnx::Conv_746[FLOAT, 336x112x1x1]
%onnx::Conv_747[FLOAT, 336]
%onnx::Conv_749[FLOAT, 336x1x5x5]
%onnx::Conv_752[FLOAT, 112x336x1x1]
%onnx::Conv_755[FLOAT, 336x112x1x1]
%onnx::Conv_758[FLOAT, 336x1x3x3]
%onnx::Conv_761[FLOAT, 112x336x1x1]
%onnx::Conv_764[FLOAT, 336x112x1x1]
%onnx::Conv_767[FLOAT, 336x1x3x3]
%onnx::Conv_770[FLOAT, 184x336x1x1]
%onnx::Conv_771[FLOAT, 184]
%onnx::Conv_773[FLOAT, 184x184x1x1]
%onnx::Conv_776[FLOAT, 184x1x3x3]
%onnx::Conv_779[FLOAT, 184x184x1x1]
%onnx::Conv_782[FLOAT, 1104x184x1x1]
%onnx::Conv_783[FLOAT, 1104]
%onnx::Conv_785[FLOAT, 1104x1x5x5]
%onnx::Conv_788[FLOAT, 184x1104x1x1]
%onnx::Conv_791[FLOAT, 184x92x1x1]
%onnx::Conv_794[FLOAT, 184x1x3x3]
%onnx::Conv_797[FLOAT, 352x92x1x1]
%onnx::Conv_798[FLOAT, 352]
%onnx::Conv_800[FLOAT, 1504x352x1x1]
%onnx::Conv_801[FLOAT, 1504]
) {
%onnx::Conv_795 = Identity(%onnx::Conv_771)
%onnx::Conv_792 = Identity(%onnx::Conv_771)
%onnx::Conv_789 = Identity(%onnx::Conv_771)
%onnx::Conv_786 = Identity(%onnx::Conv_783)
%onnx::Conv_780 = Identity(%onnx::Conv_771)
%onnx::Conv_777 = Identity(%onnx::Conv_771)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_768 = Identity(%onnx::Conv_747)
%onnx::Conv_765 = Identity(%onnx::Conv_747)
%onnx::Conv_762 = Identity(%onnx::Conv_735)
%onnx::Conv_759 = Identity(%onnx::Conv_747)
%onnx::Conv_756 = Identity(%onnx::Conv_747)
%onnx::Conv_753 = Identity(%onnx::Conv_735)
%onnx::Conv_750 = Identity(%onnx::Conv_747)
%onnx::Conv_744 = Identity(%onnx::Conv_735)
%onnx::Conv_741 = Identity(%onnx::Conv_735)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_732 = Identity(%onnx::Conv_702)
%onnx::Conv_729 = Identity(%onnx::Conv_702)
%onnx::Conv_726 = Identity(%onnx::Conv_699)
%onnx::Conv_723 = Identity(%onnx::Conv_699)
%onnx::Conv_720 = Identity(%onnx::Conv_699)
%onnx::Conv_717 = Identity(%onnx::Conv_699)
%onnx::Conv_714 = Identity(%onnx::Conv_702)
%onnx::Conv_711 = Identity(%onnx::Conv_702)
%onnx::Conv_708 = Identity(%onnx::Conv_699)
%onnx::Conv_705 = Identity(%onnx::Conv_702)
%onnx::Conv_696 = Identity(%onnx::Conv_693)
%onnx::Conv_690 = Identity(%onnx::Conv_663)
%onnx::Conv_687 = Identity(%onnx::Conv_663)
%onnx::Conv_684 = Identity(%onnx::Conv_663)
%onnx::Conv_681 = Identity(%onnx::Conv_663)
%onnx::Conv_678 = Identity(%onnx::Conv_666)
%onnx::Conv_675 = Identity(%onnx::Conv_666)
%onnx::Conv_672 = Identity(%onnx::Conv_663)
%onnx::Conv_669 = Identity(%onnx::Conv_666)
%onnx::Conv_660 = Identity(%onnx::Conv_636)
%onnx::Conv_657 = Identity(%onnx::Conv_636)
%onnx::Conv_654 = Identity(%onnx::Conv_636)
%onnx::Conv_651 = Identity(%onnx::Conv_648)
%onnx::Conv_645 = Identity(%onnx::Conv_636)
%onnx::Conv_642 = Identity(%onnx::Conv_636)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%onnx::Conv_633 = Identity(%onnx::Conv_624)
%onnx::Conv_630 = Identity(%onnx::Conv_627)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_623, %onnx::Conv_624)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.1/relu/Relu_output_0 = Relu(%/cells.1/conv/Conv_output_0)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/relu/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/relu/Relu_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%621 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %621
}
|
val_accuracy
| 0 | 64,176,768 | 1,775,164 |
{'zcp_synflow': 74.8747466327595, 'zcp_zen': 66.5299072265625, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.14085359871387482, 'zcp_flops': 64176768.0, 'zcp_grad_norm': 21.645309448242188, 'zcp_grasp': -0.060184478759765625, 'zcp_jacov': -16.05609081140185, 'zcp_l2_norm': 618.017578125, 'zcp_nwot': 209.75088031533096, 'zcp_params': 1775164.0, 'zcp_plain': -0.002521165646612644, 'zcp_snip': 42.30610656738281, 'lat_1080ti_1': 0.457186655303293, 'lat_1080ti_32': 0.44442700943978336, 'lat_1080ti_64': 0.28608921212262006, 'lat_2080ti_1': 0.5287647542114394, 'lat_2080ti_32': 0.4111796978695423, 'lat_2080ti_64': 0.2836364251828129, 'lat_essential_ph_1': 0.41509433962264153, 'lat_eyeriss': 0.38144793061082266, 'lat_fpga': 0.3995487784363589, 'lat_gold_6226': 0.37824938648293144, 'lat_gold_6240': 0.5325781310720347, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.3727315765261564, 'lat_raspi4': 0.36862909401717714, 'lat_samsung_a50': 0.25263157894736843, 'lat_samsung_s7': 0.2204724409448819, 'lat_silver_4114': 0.47146975591765694, 'lat_silver_4210r': 0.5063576866301295, 'lat_titan_rtx_1': 0.509732499959986, 'lat_titan_rtx_32': 0.41942350294279324, 'lat_titan_rtx_64': 0.30568221111605137, 'lat_titanx_1': 0.2660874861490637, 'lat_titanx_32': 0.36301409950674673, 'lat_titanx_64': 0.3263726253732479, 'lat_titanxp_1': 0.4661065391354246, 'lat_titanxp_32': 0.3960277262246361, 'lat_titanxp_64': 0.2825424515872569}
| |
FBNet_4578
|
FBNet
|
4578
|
4578
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_640[FLOAT, 16x3x3x3]
%onnx::Conv_641[FLOAT, 16]
%onnx::Conv_643[FLOAT, 16x16x1x1]
%onnx::Conv_646[FLOAT, 16x1x5x5]
%onnx::Conv_649[FLOAT, 16x16x1x1]
%onnx::Conv_652[FLOAT, 48x16x1x1]
%onnx::Conv_653[FLOAT, 48]
%onnx::Conv_655[FLOAT, 48x1x3x3]
%onnx::Conv_658[FLOAT, 24x48x1x1]
%onnx::Conv_659[FLOAT, 24]
%onnx::Conv_661[FLOAT, 24x12x1x1]
%onnx::Conv_664[FLOAT, 24x1x3x3]
%onnx::Conv_667[FLOAT, 24x12x1x1]
%onnx::Conv_670[FLOAT, 24x24x1x1]
%onnx::Conv_673[FLOAT, 24x1x5x5]
%onnx::Conv_676[FLOAT, 24x24x1x1]
%onnx::Conv_679[FLOAT, 144x24x1x1]
%onnx::Conv_680[FLOAT, 144]
%onnx::Conv_682[FLOAT, 144x1x3x3]
%onnx::Conv_685[FLOAT, 24x144x1x1]
%onnx::Conv_688[FLOAT, 144x24x1x1]
%onnx::Conv_691[FLOAT, 144x1x5x5]
%onnx::Conv_694[FLOAT, 32x144x1x1]
%onnx::Conv_695[FLOAT, 32]
%onnx::Conv_697[FLOAT, 192x32x1x1]
%onnx::Conv_698[FLOAT, 192]
%onnx::Conv_700[FLOAT, 192x1x5x5]
%onnx::Conv_703[FLOAT, 32x192x1x1]
%onnx::Conv_706[FLOAT, 96x32x1x1]
%onnx::Conv_707[FLOAT, 96]
%onnx::Conv_709[FLOAT, 96x1x5x5]
%onnx::Conv_712[FLOAT, 32x96x1x1]
%onnx::Conv_715[FLOAT, 96x32x1x1]
%onnx::Conv_718[FLOAT, 96x1x5x5]
%onnx::Conv_721[FLOAT, 32x96x1x1]
%onnx::Conv_724[FLOAT, 192x32x1x1]
%onnx::Conv_727[FLOAT, 192x1x3x3]
%onnx::Conv_730[FLOAT, 64x192x1x1]
%onnx::Conv_731[FLOAT, 64]
%onnx::Conv_733[FLOAT, 192x64x1x1]
%onnx::Conv_736[FLOAT, 192x1x5x5]
%onnx::Conv_739[FLOAT, 64x192x1x1]
%onnx::Conv_742[FLOAT, 64x64x1x1]
%onnx::Conv_745[FLOAT, 64x1x3x3]
%onnx::Conv_748[FLOAT, 64x64x1x1]
%onnx::Conv_751[FLOAT, 192x64x1x1]
%onnx::Conv_754[FLOAT, 192x1x3x3]
%onnx::Conv_757[FLOAT, 112x192x1x1]
%onnx::Conv_758[FLOAT, 112]
%onnx::Conv_760[FLOAT, 672x112x1x1]
%onnx::Conv_761[FLOAT, 672]
%onnx::Conv_763[FLOAT, 672x1x3x3]
%onnx::Conv_766[FLOAT, 112x672x1x1]
%onnx::Conv_769[FLOAT, 112x56x1x1]
%onnx::Conv_772[FLOAT, 112x1x3x3]
%onnx::Conv_775[FLOAT, 112x56x1x1]
%onnx::Conv_778[FLOAT, 112x112x1x1]
%onnx::Conv_781[FLOAT, 112x1x3x3]
%onnx::Conv_784[FLOAT, 184x112x1x1]
%onnx::Conv_785[FLOAT, 184]
%onnx::Conv_787[FLOAT, 1104x184x1x1]
%onnx::Conv_788[FLOAT, 1104]
%onnx::Conv_790[FLOAT, 1104x1x3x3]
%onnx::Conv_793[FLOAT, 184x1104x1x1]
%onnx::Conv_796[FLOAT, 184x92x1x1]
%onnx::Conv_799[FLOAT, 184x1x3x3]
%onnx::Conv_802[FLOAT, 184x92x1x1]
%onnx::Conv_805[FLOAT, 184x92x1x1]
%onnx::Conv_808[FLOAT, 184x1x3x3]
%onnx::Conv_811[FLOAT, 184x92x1x1]
%onnx::Conv_814[FLOAT, 1104x184x1x1]
%onnx::Conv_817[FLOAT, 1104x1x5x5]
%onnx::Conv_820[FLOAT, 352x1104x1x1]
%onnx::Conv_821[FLOAT, 352]
%onnx::Conv_823[FLOAT, 1504x352x1x1]
%onnx::Conv_824[FLOAT, 1504]
) {
%onnx::Conv_818 = Identity(%onnx::Conv_788)
%onnx::Conv_815 = Identity(%onnx::Conv_788)
%onnx::Conv_812 = Identity(%onnx::Conv_785)
%onnx::Conv_809 = Identity(%onnx::Conv_785)
%onnx::Conv_806 = Identity(%onnx::Conv_785)
%onnx::Conv_803 = Identity(%onnx::Conv_785)
%onnx::Conv_800 = Identity(%onnx::Conv_785)
%onnx::Conv_797 = Identity(%onnx::Conv_785)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_788)
%onnx::Conv_782 = Identity(%onnx::Conv_758)
%onnx::Conv_779 = Identity(%onnx::Conv_758)
%onnx::Conv_776 = Identity(%onnx::Conv_758)
%onnx::Conv_773 = Identity(%onnx::Conv_758)
%onnx::Conv_770 = Identity(%onnx::Conv_758)
%onnx::Conv_767 = Identity(%onnx::Conv_758)
%onnx::Conv_764 = Identity(%onnx::Conv_761)
%onnx::Conv_755 = Identity(%onnx::Conv_698)
%onnx::Conv_752 = Identity(%onnx::Conv_698)
%onnx::Conv_749 = Identity(%onnx::Conv_731)
%onnx::Conv_746 = Identity(%onnx::Conv_731)
%onnx::Conv_743 = Identity(%onnx::Conv_731)
%onnx::Conv_740 = Identity(%onnx::Conv_731)
%onnx::Conv_737 = Identity(%onnx::Conv_698)
%onnx::Conv_734 = Identity(%onnx::Conv_698)
%onnx::Conv_728 = Identity(%onnx::Conv_698)
%onnx::Conv_725 = Identity(%onnx::Conv_698)
%onnx::Conv_722 = Identity(%onnx::Conv_695)
%onnx::Conv_719 = Identity(%onnx::Conv_707)
%onnx::Conv_716 = Identity(%onnx::Conv_707)
%onnx::Conv_713 = Identity(%onnx::Conv_695)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_695)
%onnx::Conv_701 = Identity(%onnx::Conv_698)
%onnx::Conv_692 = Identity(%onnx::Conv_680)
%onnx::Conv_689 = Identity(%onnx::Conv_680)
%onnx::Conv_686 = Identity(%onnx::Conv_659)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_677 = Identity(%onnx::Conv_659)
%onnx::Conv_674 = Identity(%onnx::Conv_659)
%onnx::Conv_671 = Identity(%onnx::Conv_659)
%onnx::Conv_668 = Identity(%onnx::Conv_659)
%onnx::Conv_665 = Identity(%onnx::Conv_659)
%onnx::Conv_662 = Identity(%onnx::Conv_659)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_650 = Identity(%onnx::Conv_641)
%onnx::Conv_647 = Identity(%onnx::Conv_641)
%onnx::Conv_644 = Identity(%onnx::Conv_641)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_640, %onnx::Conv_641)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%638 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %638
}
|
val_accuracy
| 0 | 78,237,056 | 2,171,220 |
{'zcp_synflow': 76.01797792332857, 'zcp_zen': 67.53538513183594, 'zcp_epe_nas': 20.419887069448357, 'zcp_fisher': 0.0886489525437355, 'zcp_flops': 78237056.0, 'zcp_grad_norm': 26.621566772460938, 'zcp_grasp': 0.00148773193359375, 'zcp_jacov': -16.07186613952858, 'zcp_l2_norm': 628.4697265625, 'zcp_nwot': 215.3433305065159, 'zcp_params': 2171220.0, 'zcp_plain': -0.006181579548865557, 'zcp_snip': 44.961734771728516, 'lat_1080ti_1': 0.6480620236040147, 'lat_1080ti_32': 0.5480611556128132, 'lat_1080ti_64': 0.4927688092640396, 'lat_2080ti_1': 0.6225701586408326, 'lat_2080ti_32': 0.5561980052487123, 'lat_2080ti_64': 0.524904614716212, 'lat_essential_ph_1': 0.37735849056603776, 'lat_eyeriss': 0.5535455595694352, 'lat_fpga': 0.5836986314924943, 'lat_gold_6226': 0.4355251738455946, 'lat_gold_6240': 0.6015694579791486, 'lat_pixel2': 0.3695652173913043, 'lat_pixel3': 0.533047850731961, 'lat_raspi4': 0.6455076644564013, 'lat_samsung_a50': 0.23157894736842105, 'lat_samsung_s7': 0.2047244094488189, 'lat_silver_4114': 0.5919620899316599, 'lat_silver_4210r': 0.6068770428576674, 'lat_titan_rtx_1': 0.5709848225211787, 'lat_titan_rtx_32': 0.51752512832387, 'lat_titan_rtx_64': 0.5194940412564165, 'lat_titanx_1': 0.32287654779925695, 'lat_titanx_32': 0.5013894851062123, 'lat_titanx_64': 0.4859891970294691, 'lat_titanxp_1': 0.5497579436716061, 'lat_titanxp_32': 0.5256750671253023, 'lat_titanxp_64': 0.5020953258858022}
| |
FBNet_175
|
FBNet
|
175
|
175
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_659[FLOAT, 16x3x3x3]
%onnx::Conv_660[FLOAT, 16]
%onnx::Conv_662[FLOAT, 16x16x1x1]
%onnx::Conv_665[FLOAT, 16x1x5x5]
%onnx::Conv_668[FLOAT, 16x16x1x1]
%onnx::Conv_671[FLOAT, 16x16x1x1]
%onnx::Conv_674[FLOAT, 16x1x3x3]
%onnx::Conv_677[FLOAT, 24x16x1x1]
%onnx::Conv_678[FLOAT, 24]
%onnx::Conv_680[FLOAT, 24x12x1x1]
%onnx::Conv_683[FLOAT, 24x1x3x3]
%onnx::Conv_686[FLOAT, 24x12x1x1]
%onnx::Conv_689[FLOAT, 72x24x1x1]
%onnx::Conv_690[FLOAT, 72]
%onnx::Conv_692[FLOAT, 72x1x3x3]
%onnx::Conv_695[FLOAT, 24x72x1x1]
%onnx::Conv_698[FLOAT, 24x24x1x1]
%onnx::Conv_701[FLOAT, 24x1x3x3]
%onnx::Conv_704[FLOAT, 32x24x1x1]
%onnx::Conv_705[FLOAT, 32]
%onnx::Conv_707[FLOAT, 96x32x1x1]
%onnx::Conv_708[FLOAT, 96]
%onnx::Conv_710[FLOAT, 96x1x5x5]
%onnx::Conv_713[FLOAT, 32x96x1x1]
%onnx::Conv_716[FLOAT, 32x32x1x1]
%onnx::Conv_719[FLOAT, 32x1x5x5]
%onnx::Conv_722[FLOAT, 32x32x1x1]
%onnx::Conv_725[FLOAT, 32x16x1x1]
%onnx::Conv_728[FLOAT, 32x1x5x5]
%onnx::Conv_731[FLOAT, 32x16x1x1]
%onnx::Conv_734[FLOAT, 192x32x1x1]
%onnx::Conv_735[FLOAT, 192]
%onnx::Conv_737[FLOAT, 192x1x5x5]
%onnx::Conv_740[FLOAT, 64x192x1x1]
%onnx::Conv_741[FLOAT, 64]
%onnx::Conv_743[FLOAT, 192x64x1x1]
%onnx::Conv_746[FLOAT, 192x1x5x5]
%onnx::Conv_749[FLOAT, 64x192x1x1]
%onnx::Conv_752[FLOAT, 64x32x1x1]
%onnx::Conv_755[FLOAT, 64x1x3x3]
%onnx::Conv_758[FLOAT, 64x32x1x1]
%onnx::Conv_761[FLOAT, 384x64x1x1]
%onnx::Conv_762[FLOAT, 384]
%onnx::Conv_764[FLOAT, 384x1x5x5]
%onnx::Conv_767[FLOAT, 112x384x1x1]
%onnx::Conv_768[FLOAT, 112]
%onnx::Conv_770[FLOAT, 672x112x1x1]
%onnx::Conv_771[FLOAT, 672]
%onnx::Conv_773[FLOAT, 672x1x3x3]
%onnx::Conv_776[FLOAT, 112x672x1x1]
%onnx::Conv_779[FLOAT, 336x112x1x1]
%onnx::Conv_780[FLOAT, 336]
%onnx::Conv_782[FLOAT, 336x1x3x3]
%onnx::Conv_785[FLOAT, 112x336x1x1]
%onnx::Conv_788[FLOAT, 112x112x1x1]
%onnx::Conv_791[FLOAT, 112x1x3x3]
%onnx::Conv_794[FLOAT, 112x112x1x1]
%onnx::Conv_797[FLOAT, 672x112x1x1]
%onnx::Conv_800[FLOAT, 672x1x5x5]
%onnx::Conv_803[FLOAT, 184x672x1x1]
%onnx::Conv_804[FLOAT, 184]
%onnx::Conv_806[FLOAT, 1104x184x1x1]
%onnx::Conv_807[FLOAT, 1104]
%onnx::Conv_809[FLOAT, 1104x1x5x5]
%onnx::Conv_812[FLOAT, 184x1104x1x1]
%onnx::Conv_815[FLOAT, 184x92x1x1]
%onnx::Conv_818[FLOAT, 184x1x5x5]
%onnx::Conv_821[FLOAT, 184x92x1x1]
%onnx::Conv_824[FLOAT, 184x92x1x1]
%onnx::Conv_827[FLOAT, 184x1x5x5]
%onnx::Conv_830[FLOAT, 184x92x1x1]
%onnx::Conv_833[FLOAT, 184x184x1x1]
%onnx::Conv_836[FLOAT, 184x1x5x5]
%onnx::Conv_839[FLOAT, 352x184x1x1]
%onnx::Conv_840[FLOAT, 352]
%onnx::Conv_842[FLOAT, 1504x352x1x1]
%onnx::Conv_843[FLOAT, 1504]
) {
%onnx::Conv_837 = Identity(%onnx::Conv_804)
%onnx::Conv_834 = Identity(%onnx::Conv_804)
%onnx::Conv_831 = Identity(%onnx::Conv_804)
%onnx::Conv_828 = Identity(%onnx::Conv_804)
%onnx::Conv_825 = Identity(%onnx::Conv_804)
%onnx::Conv_822 = Identity(%onnx::Conv_804)
%onnx::Conv_819 = Identity(%onnx::Conv_804)
%onnx::Conv_816 = Identity(%onnx::Conv_804)
%onnx::Conv_813 = Identity(%onnx::Conv_804)
%onnx::Conv_810 = Identity(%onnx::Conv_807)
%onnx::Conv_801 = Identity(%onnx::Conv_771)
%onnx::Conv_798 = Identity(%onnx::Conv_771)
%onnx::Conv_795 = Identity(%onnx::Conv_768)
%onnx::Conv_792 = Identity(%onnx::Conv_768)
%onnx::Conv_789 = Identity(%onnx::Conv_768)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_765 = Identity(%onnx::Conv_762)
%onnx::Conv_759 = Identity(%onnx::Conv_741)
%onnx::Conv_756 = Identity(%onnx::Conv_741)
%onnx::Conv_753 = Identity(%onnx::Conv_741)
%onnx::Conv_750 = Identity(%onnx::Conv_741)
%onnx::Conv_747 = Identity(%onnx::Conv_735)
%onnx::Conv_744 = Identity(%onnx::Conv_735)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_732 = Identity(%onnx::Conv_705)
%onnx::Conv_729 = Identity(%onnx::Conv_705)
%onnx::Conv_726 = Identity(%onnx::Conv_705)
%onnx::Conv_723 = Identity(%onnx::Conv_705)
%onnx::Conv_720 = Identity(%onnx::Conv_705)
%onnx::Conv_717 = Identity(%onnx::Conv_705)
%onnx::Conv_714 = Identity(%onnx::Conv_705)
%onnx::Conv_711 = Identity(%onnx::Conv_708)
%onnx::Conv_702 = Identity(%onnx::Conv_678)
%onnx::Conv_699 = Identity(%onnx::Conv_678)
%onnx::Conv_696 = Identity(%onnx::Conv_678)
%onnx::Conv_693 = Identity(%onnx::Conv_690)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_678)
%onnx::Conv_681 = Identity(%onnx::Conv_678)
%onnx::Conv_675 = Identity(%onnx::Conv_660)
%onnx::Conv_672 = Identity(%onnx::Conv_660)
%onnx::Conv_669 = Identity(%onnx::Conv_660)
%onnx::Conv_666 = Identity(%onnx::Conv_660)
%onnx::Conv_663 = Identity(%onnx::Conv_660)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_659, %onnx::Conv_660)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%657 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %657
}
|
val_accuracy
| 0 | 65,798,272 | 1,952,148 |
{'zcp_synflow': 76.00557843445637, 'zcp_zen': 67.30181884765625, 'zcp_epe_nas': 10.149686733669459, 'zcp_fisher': 0.1172313466668129, 'zcp_flops': 65798272.0, 'zcp_grad_norm': 23.921466827392578, 'zcp_grasp': -0.014698028564453125, 'zcp_jacov': -16.038728595704438, 'zcp_l2_norm': 618.9283447265625, 'zcp_nwot': 207.88304754708068, 'zcp_params': 1952148.0, 'zcp_plain': 0.00971683207899332, 'zcp_snip': 38.680747985839844, 'lat_1080ti_1': 0.5404450798241289, 'lat_1080ti_32': 0.42092763630957586, 'lat_1080ti_64': 0.2552493212303485, 'lat_2080ti_1': 0.6931106272710047, 'lat_2080ti_32': 0.44590243091893267, 'lat_2080ti_64': 0.2696598687194483, 'lat_essential_ph_1': 0.4528301886792453, 'lat_eyeriss': 0.358388498707205, 'lat_fpga': 0.41768669786956125, 'lat_gold_6226': 0.3634666008564193, 'lat_gold_6240': 0.5187070807324338, 'lat_pixel2': 0.30434782608695654, 'lat_pixel3': 0.34230341659327546, 'lat_raspi4': 0.3557460017203072, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.2204724409448819, 'lat_silver_4114': 0.5395269482511827, 'lat_silver_4210r': 0.552635723908698, 'lat_titan_rtx_1': 0.5754963114938899, 'lat_titan_rtx_32': 0.46592805253333514, 'lat_titan_rtx_64': 0.3150114174556089, 'lat_titanx_1': 0.30545893159670356, 'lat_titanx_32': 0.3733386529459643, 'lat_titanx_64': 0.25741955539901046, 'lat_titanxp_1': 0.5353343678681108, 'lat_titanxp_32': 0.42920946213837396, 'lat_titanxp_64': 0.2730421671189178}
| |
FBNet_214
|
FBNet
|
214
|
214
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_678[FLOAT, 16x3x3x3]
%onnx::Conv_679[FLOAT, 16]
%onnx::Conv_681[FLOAT, 96x16x1x1]
%onnx::Conv_682[FLOAT, 96]
%onnx::Conv_684[FLOAT, 96x1x3x3]
%onnx::Conv_687[FLOAT, 16x96x1x1]
%onnx::Conv_690[FLOAT, 24x16x1x1]
%onnx::Conv_691[FLOAT, 24]
%onnx::Conv_693[FLOAT, 24x12x1x1]
%onnx::Conv_696[FLOAT, 24x1x3x3]
%onnx::Conv_699[FLOAT, 24x12x1x1]
%onnx::Conv_702[FLOAT, 24x24x1x1]
%onnx::Conv_705[FLOAT, 24x1x5x5]
%onnx::Conv_708[FLOAT, 24x24x1x1]
%onnx::Conv_711[FLOAT, 24x24x1x1]
%onnx::Conv_714[FLOAT, 24x1x3x3]
%onnx::Conv_717[FLOAT, 24x24x1x1]
%onnx::Conv_720[FLOAT, 72x24x1x1]
%onnx::Conv_721[FLOAT, 72]
%onnx::Conv_723[FLOAT, 72x1x5x5]
%onnx::Conv_726[FLOAT, 32x72x1x1]
%onnx::Conv_727[FLOAT, 32]
%onnx::Conv_729[FLOAT, 192x32x1x1]
%onnx::Conv_730[FLOAT, 192]
%onnx::Conv_732[FLOAT, 192x1x5x5]
%onnx::Conv_735[FLOAT, 32x192x1x1]
%onnx::Conv_738[FLOAT, 192x32x1x1]
%onnx::Conv_741[FLOAT, 192x1x5x5]
%onnx::Conv_744[FLOAT, 32x192x1x1]
%onnx::Conv_747[FLOAT, 32x16x1x1]
%onnx::Conv_750[FLOAT, 32x1x3x3]
%onnx::Conv_753[FLOAT, 32x16x1x1]
%onnx::Conv_756[FLOAT, 192x32x1x1]
%onnx::Conv_759[FLOAT, 192x1x3x3]
%onnx::Conv_762[FLOAT, 64x192x1x1]
%onnx::Conv_763[FLOAT, 64]
%onnx::Conv_765[FLOAT, 384x64x1x1]
%onnx::Conv_766[FLOAT, 384]
%onnx::Conv_768[FLOAT, 384x1x3x3]
%onnx::Conv_771[FLOAT, 64x384x1x1]
%onnx::Conv_774[FLOAT, 384x64x1x1]
%onnx::Conv_777[FLOAT, 384x1x3x3]
%onnx::Conv_780[FLOAT, 64x384x1x1]
%onnx::Conv_783[FLOAT, 384x64x1x1]
%onnx::Conv_786[FLOAT, 384x1x3x3]
%onnx::Conv_789[FLOAT, 64x384x1x1]
%onnx::Conv_792[FLOAT, 112x64x1x1]
%onnx::Conv_793[FLOAT, 112]
%onnx::Conv_795[FLOAT, 112x56x1x1]
%onnx::Conv_798[FLOAT, 112x1x3x3]
%onnx::Conv_801[FLOAT, 112x56x1x1]
%onnx::Conv_804[FLOAT, 112x56x1x1]
%onnx::Conv_807[FLOAT, 112x1x5x5]
%onnx::Conv_810[FLOAT, 112x56x1x1]
%onnx::Conv_813[FLOAT, 112x56x1x1]
%onnx::Conv_816[FLOAT, 112x1x3x3]
%onnx::Conv_819[FLOAT, 112x56x1x1]
%onnx::Conv_822[FLOAT, 112x112x1x1]
%onnx::Conv_825[FLOAT, 112x1x3x3]
%onnx::Conv_828[FLOAT, 184x112x1x1]
%onnx::Conv_829[FLOAT, 184]
%onnx::Conv_831[FLOAT, 184x184x1x1]
%onnx::Conv_834[FLOAT, 184x1x5x5]
%onnx::Conv_837[FLOAT, 184x184x1x1]
%onnx::Conv_840[FLOAT, 1104x184x1x1]
%onnx::Conv_841[FLOAT, 1104]
%onnx::Conv_843[FLOAT, 1104x1x3x3]
%onnx::Conv_846[FLOAT, 184x1104x1x1]
%onnx::Conv_849[FLOAT, 552x184x1x1]
%onnx::Conv_850[FLOAT, 552]
%onnx::Conv_852[FLOAT, 552x1x5x5]
%onnx::Conv_855[FLOAT, 184x552x1x1]
%onnx::Conv_858[FLOAT, 184x184x1x1]
%onnx::Conv_861[FLOAT, 184x1x5x5]
%onnx::Conv_864[FLOAT, 352x184x1x1]
%onnx::Conv_865[FLOAT, 352]
%onnx::Conv_867[FLOAT, 1504x352x1x1]
%onnx::Conv_868[FLOAT, 1504]
) {
%onnx::Conv_862 = Identity(%onnx::Conv_829)
%onnx::Conv_859 = Identity(%onnx::Conv_829)
%onnx::Conv_856 = Identity(%onnx::Conv_829)
%onnx::Conv_853 = Identity(%onnx::Conv_850)
%onnx::Conv_847 = Identity(%onnx::Conv_829)
%onnx::Conv_844 = Identity(%onnx::Conv_841)
%onnx::Conv_838 = Identity(%onnx::Conv_829)
%onnx::Conv_835 = Identity(%onnx::Conv_829)
%onnx::Conv_832 = Identity(%onnx::Conv_829)
%onnx::Conv_826 = Identity(%onnx::Conv_793)
%onnx::Conv_823 = Identity(%onnx::Conv_793)
%onnx::Conv_820 = Identity(%onnx::Conv_793)
%onnx::Conv_817 = Identity(%onnx::Conv_793)
%onnx::Conv_814 = Identity(%onnx::Conv_793)
%onnx::Conv_811 = Identity(%onnx::Conv_793)
%onnx::Conv_808 = Identity(%onnx::Conv_793)
%onnx::Conv_805 = Identity(%onnx::Conv_793)
%onnx::Conv_802 = Identity(%onnx::Conv_793)
%onnx::Conv_799 = Identity(%onnx::Conv_793)
%onnx::Conv_796 = Identity(%onnx::Conv_793)
%onnx::Conv_790 = Identity(%onnx::Conv_763)
%onnx::Conv_787 = Identity(%onnx::Conv_766)
%onnx::Conv_784 = Identity(%onnx::Conv_766)
%onnx::Conv_781 = Identity(%onnx::Conv_763)
%onnx::Conv_778 = Identity(%onnx::Conv_766)
%onnx::Conv_775 = Identity(%onnx::Conv_766)
%onnx::Conv_772 = Identity(%onnx::Conv_763)
%onnx::Conv_769 = Identity(%onnx::Conv_766)
%onnx::Conv_760 = Identity(%onnx::Conv_730)
%onnx::Conv_757 = Identity(%onnx::Conv_730)
%onnx::Conv_754 = Identity(%onnx::Conv_727)
%onnx::Conv_751 = Identity(%onnx::Conv_727)
%onnx::Conv_748 = Identity(%onnx::Conv_727)
%onnx::Conv_745 = Identity(%onnx::Conv_727)
%onnx::Conv_742 = Identity(%onnx::Conv_730)
%onnx::Conv_739 = Identity(%onnx::Conv_730)
%onnx::Conv_736 = Identity(%onnx::Conv_727)
%onnx::Conv_733 = Identity(%onnx::Conv_730)
%onnx::Conv_724 = Identity(%onnx::Conv_721)
%onnx::Conv_718 = Identity(%onnx::Conv_691)
%onnx::Conv_715 = Identity(%onnx::Conv_691)
%onnx::Conv_712 = Identity(%onnx::Conv_691)
%onnx::Conv_709 = Identity(%onnx::Conv_691)
%onnx::Conv_706 = Identity(%onnx::Conv_691)
%onnx::Conv_703 = Identity(%onnx::Conv_691)
%onnx::Conv_700 = Identity(%onnx::Conv_691)
%onnx::Conv_697 = Identity(%onnx::Conv_691)
%onnx::Conv_694 = Identity(%onnx::Conv_691)
%onnx::Conv_688 = Identity(%onnx::Conv_679)
%onnx::Conv_685 = Identity(%onnx::Conv_682)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_678, %onnx::Conv_679)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.1/relu/Relu_output_0 = Relu(%/cells.1/conv/Conv_output_0)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/relu/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/relu/Relu_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%676 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %676
}
|
val_accuracy
| 0 | 62,216,832 | 1,825,324 |
{'zcp_synflow': 78.9042167677412, 'zcp_zen': 69.7698974609375, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.20022794604301453, 'zcp_flops': 62216832.0, 'zcp_grad_norm': 26.06709861755371, 'zcp_grasp': -0.08728790283203125, 'zcp_jacov': -16.06581758833258, 'zcp_l2_norm': 649.5956420898438, 'zcp_nwot': 212.4082183755881, 'zcp_params': 1825324.0, 'zcp_plain': -0.004324276465922594, 'zcp_snip': 52.3967399597168, 'lat_1080ti_1': 0.5844597797166337, 'lat_1080ti_32': 0.6121852906050901, 'lat_1080ti_64': 0.37993446483289106, 'lat_2080ti_1': 0.6847957409440705, 'lat_2080ti_32': 0.5356846946733275, 'lat_2080ti_64': 0.415023068475492, 'lat_essential_ph_1': 0.2641509433962264, 'lat_eyeriss': 0.43771661195973094, 'lat_fpga': 0.3718224007757843, 'lat_gold_6226': 0.3450401630265633, 'lat_gold_6240': 0.5758281133390213, 'lat_pixel2': 0.30434782608695654, 'lat_pixel3': 0.37087149617509174, 'lat_raspi4': 0.38225647810447805, 'lat_samsung_a50': 0.18947368421052632, 'lat_samsung_s7': 0.18110236220472442, 'lat_silver_4114': 0.583069600161751, 'lat_silver_4210r': 0.6421043842035478, 'lat_titan_rtx_1': 0.65368401289512, 'lat_titan_rtx_32': 0.5532585647981915, 'lat_titan_rtx_64': 0.453779970465346, 'lat_titanx_1': 0.34988230295151884, 'lat_titanx_32': 0.4870205148315295, 'lat_titanx_64': 0.38860031510379106, 'lat_titanxp_1': 0.607742337235571, 'lat_titanxp_32': 0.5133847497546249, 'lat_titanxp_64': 0.41688302036029334}
| |
FBNet_4394
|
FBNet
|
4394
|
4394
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_749[FLOAT, 16x3x3x3]
%onnx::Conv_750[FLOAT, 16]
%onnx::Conv_752[FLOAT, 16x16x1x1]
%onnx::Conv_755[FLOAT, 16x1x5x5]
%onnx::Conv_758[FLOAT, 16x16x1x1]
%onnx::Conv_761[FLOAT, 16x8x1x1]
%onnx::Conv_764[FLOAT, 16x1x3x3]
%onnx::Conv_767[FLOAT, 24x8x1x1]
%onnx::Conv_768[FLOAT, 24]
%onnx::Conv_770[FLOAT, 24x12x1x1]
%onnx::Conv_773[FLOAT, 24x1x3x3]
%onnx::Conv_776[FLOAT, 24x12x1x1]
%onnx::Conv_779[FLOAT, 24x12x1x1]
%onnx::Conv_782[FLOAT, 24x1x5x5]
%onnx::Conv_785[FLOAT, 24x12x1x1]
%onnx::Conv_788[FLOAT, 24x12x1x1]
%onnx::Conv_791[FLOAT, 24x1x3x3]
%onnx::Conv_794[FLOAT, 24x12x1x1]
%onnx::Conv_797[FLOAT, 24x24x1x1]
%onnx::Conv_800[FLOAT, 24x1x5x5]
%onnx::Conv_803[FLOAT, 32x24x1x1]
%onnx::Conv_804[FLOAT, 32]
%onnx::Conv_806[FLOAT, 32x32x1x1]
%onnx::Conv_809[FLOAT, 32x1x3x3]
%onnx::Conv_812[FLOAT, 32x32x1x1]
%onnx::Conv_815[FLOAT, 96x32x1x1]
%onnx::Conv_816[FLOAT, 96]
%onnx::Conv_818[FLOAT, 96x1x5x5]
%onnx::Conv_821[FLOAT, 32x96x1x1]
%onnx::Conv_824[FLOAT, 96x32x1x1]
%onnx::Conv_827[FLOAT, 96x1x5x5]
%onnx::Conv_830[FLOAT, 32x96x1x1]
%onnx::Conv_833[FLOAT, 32x32x1x1]
%onnx::Conv_836[FLOAT, 32x1x3x3]
%onnx::Conv_839[FLOAT, 64x32x1x1]
%onnx::Conv_840[FLOAT, 64]
%onnx::Conv_842[FLOAT, 384x64x1x1]
%onnx::Conv_843[FLOAT, 384]
%onnx::Conv_845[FLOAT, 384x1x5x5]
%onnx::Conv_848[FLOAT, 64x384x1x1]
%onnx::Conv_851[FLOAT, 64x32x1x1]
%onnx::Conv_854[FLOAT, 64x1x3x3]
%onnx::Conv_857[FLOAT, 64x32x1x1]
%onnx::Conv_860[FLOAT, 64x64x1x1]
%onnx::Conv_863[FLOAT, 64x1x3x3]
%onnx::Conv_866[FLOAT, 64x64x1x1]
%onnx::Conv_869[FLOAT, 64x32x1x1]
%onnx::Conv_872[FLOAT, 64x1x3x3]
%onnx::Conv_875[FLOAT, 112x32x1x1]
%onnx::Conv_876[FLOAT, 112]
%onnx::Conv_878[FLOAT, 336x112x1x1]
%onnx::Conv_879[FLOAT, 336]
%onnx::Conv_881[FLOAT, 336x1x3x3]
%onnx::Conv_884[FLOAT, 112x336x1x1]
%onnx::Conv_887[FLOAT, 112x56x1x1]
%onnx::Conv_890[FLOAT, 112x1x5x5]
%onnx::Conv_893[FLOAT, 112x56x1x1]
%onnx::Conv_896[FLOAT, 336x112x1x1]
%onnx::Conv_899[FLOAT, 336x1x5x5]
%onnx::Conv_902[FLOAT, 112x336x1x1]
%onnx::Conv_905[FLOAT, 336x112x1x1]
%onnx::Conv_908[FLOAT, 336x1x5x5]
%onnx::Conv_911[FLOAT, 184x336x1x1]
%onnx::Conv_912[FLOAT, 184]
%onnx::Conv_914[FLOAT, 1104x184x1x1]
%onnx::Conv_915[FLOAT, 1104]
%onnx::Conv_917[FLOAT, 1104x1x3x3]
%onnx::Conv_920[FLOAT, 184x1104x1x1]
%onnx::Conv_923[FLOAT, 1104x184x1x1]
%onnx::Conv_926[FLOAT, 1104x1x5x5]
%onnx::Conv_929[FLOAT, 184x1104x1x1]
%onnx::Conv_932[FLOAT, 1104x184x1x1]
%onnx::Conv_935[FLOAT, 1104x1x3x3]
%onnx::Conv_938[FLOAT, 184x1104x1x1]
%onnx::Conv_941[FLOAT, 184x184x1x1]
%onnx::Conv_944[FLOAT, 184x1x3x3]
%onnx::Conv_947[FLOAT, 352x184x1x1]
%onnx::Conv_948[FLOAT, 352]
%onnx::Conv_950[FLOAT, 1504x352x1x1]
%onnx::Conv_951[FLOAT, 1504]
) {
%onnx::Conv_945 = Identity(%onnx::Conv_912)
%onnx::Conv_942 = Identity(%onnx::Conv_912)
%onnx::Conv_939 = Identity(%onnx::Conv_912)
%onnx::Conv_936 = Identity(%onnx::Conv_915)
%onnx::Conv_933 = Identity(%onnx::Conv_915)
%onnx::Conv_930 = Identity(%onnx::Conv_912)
%onnx::Conv_927 = Identity(%onnx::Conv_915)
%onnx::Conv_924 = Identity(%onnx::Conv_915)
%onnx::Conv_921 = Identity(%onnx::Conv_912)
%onnx::Conv_918 = Identity(%onnx::Conv_915)
%onnx::Conv_909 = Identity(%onnx::Conv_879)
%onnx::Conv_906 = Identity(%onnx::Conv_879)
%onnx::Conv_903 = Identity(%onnx::Conv_876)
%onnx::Conv_900 = Identity(%onnx::Conv_879)
%onnx::Conv_897 = Identity(%onnx::Conv_879)
%onnx::Conv_894 = Identity(%onnx::Conv_876)
%onnx::Conv_891 = Identity(%onnx::Conv_876)
%onnx::Conv_888 = Identity(%onnx::Conv_876)
%onnx::Conv_885 = Identity(%onnx::Conv_876)
%onnx::Conv_882 = Identity(%onnx::Conv_879)
%onnx::Conv_873 = Identity(%onnx::Conv_840)
%onnx::Conv_870 = Identity(%onnx::Conv_840)
%onnx::Conv_867 = Identity(%onnx::Conv_840)
%onnx::Conv_864 = Identity(%onnx::Conv_840)
%onnx::Conv_861 = Identity(%onnx::Conv_840)
%onnx::Conv_858 = Identity(%onnx::Conv_840)
%onnx::Conv_855 = Identity(%onnx::Conv_840)
%onnx::Conv_852 = Identity(%onnx::Conv_840)
%onnx::Conv_849 = Identity(%onnx::Conv_840)
%onnx::Conv_846 = Identity(%onnx::Conv_843)
%onnx::Conv_837 = Identity(%onnx::Conv_804)
%onnx::Conv_834 = Identity(%onnx::Conv_804)
%onnx::Conv_831 = Identity(%onnx::Conv_804)
%onnx::Conv_828 = Identity(%onnx::Conv_816)
%onnx::Conv_825 = Identity(%onnx::Conv_816)
%onnx::Conv_822 = Identity(%onnx::Conv_804)
%onnx::Conv_819 = Identity(%onnx::Conv_816)
%onnx::Conv_813 = Identity(%onnx::Conv_804)
%onnx::Conv_810 = Identity(%onnx::Conv_804)
%onnx::Conv_807 = Identity(%onnx::Conv_804)
%onnx::Conv_801 = Identity(%onnx::Conv_768)
%onnx::Conv_798 = Identity(%onnx::Conv_768)
%onnx::Conv_795 = Identity(%onnx::Conv_768)
%onnx::Conv_792 = Identity(%onnx::Conv_768)
%onnx::Conv_789 = Identity(%onnx::Conv_768)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_768)
%onnx::Conv_780 = Identity(%onnx::Conv_768)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_768)
%onnx::Conv_771 = Identity(%onnx::Conv_768)
%onnx::Conv_765 = Identity(%onnx::Conv_750)
%onnx::Conv_762 = Identity(%onnx::Conv_750)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_750)
%onnx::Conv_753 = Identity(%onnx::Conv_750)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_749, %onnx::Conv_750)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_872, %onnx::Conv_873)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_875, %onnx::Conv_876)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_878, %onnx::Conv_879)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_881, %onnx::Conv_882)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_884, %onnx::Conv_885)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_887, %onnx::Conv_888)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_890, %onnx::Conv_891)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_893, %onnx::Conv_894)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_896, %onnx::Conv_897)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_899, %onnx::Conv_900)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_902, %onnx::Conv_903)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_905, %onnx::Conv_906)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_908, %onnx::Conv_909)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_911, %onnx::Conv_912)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_914, %onnx::Conv_915)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_917, %onnx::Conv_918)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_920, %onnx::Conv_921)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_923, %onnx::Conv_924)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_926, %onnx::Conv_927)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_929, %onnx::Conv_930)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_932, %onnx::Conv_933)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_935, %onnx::Conv_936)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_938, %onnx::Conv_939)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_941, %onnx::Conv_942)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_944, %onnx::Conv_945)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_947, %onnx::Conv_948)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_950, %onnx::Conv_951)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%747 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %747
}
|
val_accuracy
| 0 | 63,633,024 | 2,468,732 |
{'zcp_synflow': 80.27023744486135, 'zcp_zen': 71.53919982910156, 'zcp_epe_nas': 21.351406214568694, 'zcp_fisher': 0.20822760462760925, 'zcp_flops': 63633024.0, 'zcp_grad_norm': 25.037433624267578, 'zcp_grasp': 0.14701461791992188, 'zcp_jacov': -16.047119749193072, 'zcp_l2_norm': 673.0994873046875, 'zcp_nwot': 206.34464226381866, 'zcp_params': 2468732.0, 'zcp_plain': -0.0038637048564851284, 'zcp_snip': 44.7559814453125, 'lat_1080ti_1': 0.8130106059505023, 'lat_1080ti_32': 0.6356884448856802, 'lat_1080ti_64': 0.3588919439379145, 'lat_2080ti_1': 0.8697239279753355, 'lat_2080ti_32': 0.6965134519581263, 'lat_2080ti_64': 0.3607871304808628, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.4094951680633744, 'lat_fpga': 0.4437605755053976, 'lat_gold_6226': 0.48221840067185706, 'lat_gold_6240': 0.7924928614476048, 'lat_pixel2': 0.34782608695652173, 'lat_pixel3': 0.39808470479168223, 'lat_raspi4': 0.47844479035615306, 'lat_samsung_a50': 0.23157894736842105, 'lat_samsung_s7': 0.2204724409448819, 'lat_silver_4114': 0.8541757444492885, 'lat_silver_4210r': 0.867955499555394, 'lat_titan_rtx_1': 0.8224954683549737, 'lat_titan_rtx_32': 0.6872921796857036, 'lat_titan_rtx_64': 0.4431363707143302, 'lat_titanx_1': 0.44528802897898556, 'lat_titanx_32': 0.9079713123596841, 'lat_titanx_64': 0.35903285072319907, 'lat_titanxp_1': 0.7966351228241162, 'lat_titanxp_32': 0.6146097404280564, 'lat_titanxp_64': 0.36263616131025683}
| |
FBNet_1272
|
FBNet
|
1272
|
1272
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_650[FLOAT, 16x3x3x3]
%onnx::Conv_651[FLOAT, 16]
%onnx::Conv_653[FLOAT, 16x16x1x1]
%onnx::Conv_656[FLOAT, 16x1x3x3]
%onnx::Conv_659[FLOAT, 16x16x1x1]
%onnx::Conv_662[FLOAT, 16x16x1x1]
%onnx::Conv_665[FLOAT, 16x1x5x5]
%onnx::Conv_668[FLOAT, 24x16x1x1]
%onnx::Conv_669[FLOAT, 24]
%onnx::Conv_671[FLOAT, 24x24x1x1]
%onnx::Conv_674[FLOAT, 24x1x3x3]
%onnx::Conv_677[FLOAT, 24x24x1x1]
%onnx::Conv_680[FLOAT, 72x24x1x1]
%onnx::Conv_681[FLOAT, 72]
%onnx::Conv_683[FLOAT, 72x1x3x3]
%onnx::Conv_686[FLOAT, 24x72x1x1]
%onnx::Conv_689[FLOAT, 72x24x1x1]
%onnx::Conv_692[FLOAT, 72x1x3x3]
%onnx::Conv_695[FLOAT, 32x72x1x1]
%onnx::Conv_696[FLOAT, 32]
%onnx::Conv_698[FLOAT, 32x16x1x1]
%onnx::Conv_701[FLOAT, 32x1x5x5]
%onnx::Conv_704[FLOAT, 32x16x1x1]
%onnx::Conv_707[FLOAT, 32x16x1x1]
%onnx::Conv_710[FLOAT, 32x1x3x3]
%onnx::Conv_713[FLOAT, 32x16x1x1]
%onnx::Conv_716[FLOAT, 96x32x1x1]
%onnx::Conv_717[FLOAT, 96]
%onnx::Conv_719[FLOAT, 96x1x3x3]
%onnx::Conv_722[FLOAT, 64x96x1x1]
%onnx::Conv_723[FLOAT, 64]
%onnx::Conv_725[FLOAT, 384x64x1x1]
%onnx::Conv_726[FLOAT, 384]
%onnx::Conv_728[FLOAT, 384x1x3x3]
%onnx::Conv_731[FLOAT, 64x384x1x1]
%onnx::Conv_734[FLOAT, 64x32x1x1]
%onnx::Conv_737[FLOAT, 64x1x3x3]
%onnx::Conv_740[FLOAT, 64x32x1x1]
%onnx::Conv_743[FLOAT, 64x64x1x1]
%onnx::Conv_746[FLOAT, 64x1x3x3]
%onnx::Conv_749[FLOAT, 112x64x1x1]
%onnx::Conv_750[FLOAT, 112]
%onnx::Conv_752[FLOAT, 112x112x1x1]
%onnx::Conv_755[FLOAT, 112x1x3x3]
%onnx::Conv_758[FLOAT, 112x112x1x1]
%onnx::Conv_761[FLOAT, 112x56x1x1]
%onnx::Conv_764[FLOAT, 112x1x5x5]
%onnx::Conv_767[FLOAT, 112x56x1x1]
%onnx::Conv_770[FLOAT, 672x112x1x1]
%onnx::Conv_771[FLOAT, 672]
%onnx::Conv_773[FLOAT, 672x1x5x5]
%onnx::Conv_776[FLOAT, 112x672x1x1]
%onnx::Conv_779[FLOAT, 112x56x1x1]
%onnx::Conv_782[FLOAT, 112x1x5x5]
%onnx::Conv_785[FLOAT, 184x56x1x1]
%onnx::Conv_786[FLOAT, 184]
%onnx::Conv_788[FLOAT, 184x184x1x1]
%onnx::Conv_791[FLOAT, 184x1x3x3]
%onnx::Conv_794[FLOAT, 184x184x1x1]
%onnx::Conv_797[FLOAT, 552x184x1x1]
%onnx::Conv_798[FLOAT, 552]
%onnx::Conv_800[FLOAT, 552x1x5x5]
%onnx::Conv_803[FLOAT, 184x552x1x1]
%onnx::Conv_806[FLOAT, 184x184x1x1]
%onnx::Conv_809[FLOAT, 184x1x3x3]
%onnx::Conv_812[FLOAT, 184x184x1x1]
%onnx::Conv_815[FLOAT, 184x92x1x1]
%onnx::Conv_818[FLOAT, 184x1x3x3]
%onnx::Conv_821[FLOAT, 352x92x1x1]
%onnx::Conv_822[FLOAT, 352]
%onnx::Conv_824[FLOAT, 1504x352x1x1]
%onnx::Conv_825[FLOAT, 1504]
) {
%onnx::Conv_819 = Identity(%onnx::Conv_786)
%onnx::Conv_816 = Identity(%onnx::Conv_786)
%onnx::Conv_813 = Identity(%onnx::Conv_786)
%onnx::Conv_810 = Identity(%onnx::Conv_786)
%onnx::Conv_807 = Identity(%onnx::Conv_786)
%onnx::Conv_804 = Identity(%onnx::Conv_786)
%onnx::Conv_801 = Identity(%onnx::Conv_798)
%onnx::Conv_795 = Identity(%onnx::Conv_786)
%onnx::Conv_792 = Identity(%onnx::Conv_786)
%onnx::Conv_789 = Identity(%onnx::Conv_786)
%onnx::Conv_783 = Identity(%onnx::Conv_750)
%onnx::Conv_780 = Identity(%onnx::Conv_750)
%onnx::Conv_777 = Identity(%onnx::Conv_750)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_768 = Identity(%onnx::Conv_750)
%onnx::Conv_765 = Identity(%onnx::Conv_750)
%onnx::Conv_762 = Identity(%onnx::Conv_750)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_750)
%onnx::Conv_753 = Identity(%onnx::Conv_750)
%onnx::Conv_747 = Identity(%onnx::Conv_723)
%onnx::Conv_744 = Identity(%onnx::Conv_723)
%onnx::Conv_741 = Identity(%onnx::Conv_723)
%onnx::Conv_738 = Identity(%onnx::Conv_723)
%onnx::Conv_735 = Identity(%onnx::Conv_723)
%onnx::Conv_732 = Identity(%onnx::Conv_723)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%onnx::Conv_714 = Identity(%onnx::Conv_696)
%onnx::Conv_711 = Identity(%onnx::Conv_696)
%onnx::Conv_708 = Identity(%onnx::Conv_696)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_696)
%onnx::Conv_699 = Identity(%onnx::Conv_696)
%onnx::Conv_693 = Identity(%onnx::Conv_681)
%onnx::Conv_690 = Identity(%onnx::Conv_681)
%onnx::Conv_687 = Identity(%onnx::Conv_669)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_678 = Identity(%onnx::Conv_669)
%onnx::Conv_675 = Identity(%onnx::Conv_669)
%onnx::Conv_672 = Identity(%onnx::Conv_669)
%onnx::Conv_666 = Identity(%onnx::Conv_651)
%onnx::Conv_663 = Identity(%onnx::Conv_651)
%onnx::Conv_660 = Identity(%onnx::Conv_651)
%onnx::Conv_657 = Identity(%onnx::Conv_651)
%onnx::Conv_654 = Identity(%onnx::Conv_651)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_650, %onnx::Conv_651)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%648 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %648
}
|
val_accuracy
| 0 | 47,926,144 | 1,428,108 |
{'zcp_synflow': 69.62824914033703, 'zcp_zen': 60.576690673828125, 'zcp_epe_nas': 11.254503197619774, 'zcp_fisher': 0.08133874088525772, 'zcp_flops': 47926144.0, 'zcp_grad_norm': 17.946077346801758, 'zcp_grasp': -0.008582115173339844, 'zcp_jacov': -16.053544360083265, 'zcp_l2_norm': 533.630859375, 'zcp_nwot': 205.62829027206126, 'zcp_params': 1428108.0, 'zcp_plain': 0.0027370378375053406, 'zcp_snip': 30.57222557067871, 'lat_1080ti_1': 0.4494866260305173, 'lat_1080ti_32': 0.43399101885999414, 'lat_1080ti_64': 0.2260748650376499, 'lat_2080ti_1': 0.5195584304447107, 'lat_2080ti_32': 0.3991902074559442, 'lat_2080ti_64': 0.27119429244003557, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.16147103588652742, 'lat_fpga': 0.2154879822677841, 'lat_gold_6226': 0.15485780271389327, 'lat_gold_6240': 0.25854987568519633, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.16310272529399858, 'lat_raspi4': 0.17097460678909435, 'lat_samsung_a50': 0.08421052631578947, 'lat_samsung_s7': 0.08661417322834646, 'lat_silver_4114': 0.27924075461255876, 'lat_silver_4210r': 0.25914476864262154, 'lat_titan_rtx_1': 0.5010272604443351, 'lat_titan_rtx_32': 0.4008003658253725, 'lat_titan_rtx_64': 0.2897919084661851, 'lat_titanx_1': 0.31262007540128894, 'lat_titanx_32': 0.3266216064325949, 'lat_titanx_64': 0.2345079559315895, 'lat_titanxp_1': 0.48704758356397937, 'lat_titanxp_32': 0.3728936022025636, 'lat_titanxp_64': 0.2527171962722508}
| |
FBNet_1913
|
FBNet
|
1913
|
1913
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_669[FLOAT, 16x3x3x3]
%onnx::Conv_670[FLOAT, 16]
%onnx::Conv_672[FLOAT, 96x16x1x1]
%onnx::Conv_673[FLOAT, 96]
%onnx::Conv_675[FLOAT, 96x1x5x5]
%onnx::Conv_678[FLOAT, 16x96x1x1]
%onnx::Conv_681[FLOAT, 96x16x1x1]
%onnx::Conv_684[FLOAT, 96x1x5x5]
%onnx::Conv_687[FLOAT, 24x96x1x1]
%onnx::Conv_688[FLOAT, 24]
%onnx::Conv_690[FLOAT, 72x24x1x1]
%onnx::Conv_691[FLOAT, 72]
%onnx::Conv_693[FLOAT, 72x1x5x5]
%onnx::Conv_696[FLOAT, 24x72x1x1]
%onnx::Conv_699[FLOAT, 144x24x1x1]
%onnx::Conv_700[FLOAT, 144]
%onnx::Conv_702[FLOAT, 144x1x5x5]
%onnx::Conv_705[FLOAT, 24x144x1x1]
%onnx::Conv_708[FLOAT, 144x24x1x1]
%onnx::Conv_711[FLOAT, 144x1x3x3]
%onnx::Conv_714[FLOAT, 32x144x1x1]
%onnx::Conv_715[FLOAT, 32]
%onnx::Conv_717[FLOAT, 32x32x1x1]
%onnx::Conv_720[FLOAT, 32x1x3x3]
%onnx::Conv_723[FLOAT, 32x32x1x1]
%onnx::Conv_726[FLOAT, 32x16x1x1]
%onnx::Conv_729[FLOAT, 32x1x5x5]
%onnx::Conv_732[FLOAT, 32x16x1x1]
%onnx::Conv_735[FLOAT, 32x16x1x1]
%onnx::Conv_738[FLOAT, 32x1x3x3]
%onnx::Conv_741[FLOAT, 32x16x1x1]
%onnx::Conv_744[FLOAT, 32x32x1x1]
%onnx::Conv_747[FLOAT, 32x1x3x3]
%onnx::Conv_750[FLOAT, 64x32x1x1]
%onnx::Conv_751[FLOAT, 64]
%onnx::Conv_753[FLOAT, 384x64x1x1]
%onnx::Conv_754[FLOAT, 384]
%onnx::Conv_756[FLOAT, 384x1x3x3]
%onnx::Conv_759[FLOAT, 64x384x1x1]
%onnx::Conv_762[FLOAT, 384x64x1x1]
%onnx::Conv_765[FLOAT, 384x1x5x5]
%onnx::Conv_768[FLOAT, 64x384x1x1]
%onnx::Conv_771[FLOAT, 64x64x1x1]
%onnx::Conv_774[FLOAT, 64x1x5x5]
%onnx::Conv_777[FLOAT, 64x64x1x1]
%onnx::Conv_780[FLOAT, 64x32x1x1]
%onnx::Conv_783[FLOAT, 64x1x5x5]
%onnx::Conv_786[FLOAT, 112x32x1x1]
%onnx::Conv_787[FLOAT, 112]
%onnx::Conv_789[FLOAT, 112x112x1x1]
%onnx::Conv_792[FLOAT, 112x1x5x5]
%onnx::Conv_795[FLOAT, 112x112x1x1]
%onnx::Conv_798[FLOAT, 112x112x1x1]
%onnx::Conv_801[FLOAT, 112x1x3x3]
%onnx::Conv_804[FLOAT, 112x112x1x1]
%onnx::Conv_807[FLOAT, 112x56x1x1]
%onnx::Conv_810[FLOAT, 112x1x5x5]
%onnx::Conv_813[FLOAT, 112x56x1x1]
%onnx::Conv_816[FLOAT, 112x112x1x1]
%onnx::Conv_819[FLOAT, 112x1x3x3]
%onnx::Conv_822[FLOAT, 184x112x1x1]
%onnx::Conv_823[FLOAT, 184]
%onnx::Conv_825[FLOAT, 552x184x1x1]
%onnx::Conv_826[FLOAT, 552]
%onnx::Conv_828[FLOAT, 552x1x5x5]
%onnx::Conv_831[FLOAT, 184x552x1x1]
%onnx::Conv_834[FLOAT, 184x92x1x1]
%onnx::Conv_837[FLOAT, 184x1x3x3]
%onnx::Conv_840[FLOAT, 184x92x1x1]
%onnx::Conv_843[FLOAT, 184x184x1x1]
%onnx::Conv_846[FLOAT, 184x1x5x5]
%onnx::Conv_849[FLOAT, 184x184x1x1]
%onnx::Conv_852[FLOAT, 352x184x1x1]
%onnx::Conv_853[FLOAT, 352]
%onnx::Conv_855[FLOAT, 1504x352x1x1]
%onnx::Conv_856[FLOAT, 1504]
) {
%onnx::Conv_850 = Identity(%onnx::Conv_823)
%onnx::Conv_847 = Identity(%onnx::Conv_823)
%onnx::Conv_844 = Identity(%onnx::Conv_823)
%onnx::Conv_841 = Identity(%onnx::Conv_823)
%onnx::Conv_838 = Identity(%onnx::Conv_823)
%onnx::Conv_835 = Identity(%onnx::Conv_823)
%onnx::Conv_832 = Identity(%onnx::Conv_823)
%onnx::Conv_829 = Identity(%onnx::Conv_826)
%onnx::Conv_820 = Identity(%onnx::Conv_787)
%onnx::Conv_817 = Identity(%onnx::Conv_787)
%onnx::Conv_814 = Identity(%onnx::Conv_787)
%onnx::Conv_811 = Identity(%onnx::Conv_787)
%onnx::Conv_808 = Identity(%onnx::Conv_787)
%onnx::Conv_805 = Identity(%onnx::Conv_787)
%onnx::Conv_802 = Identity(%onnx::Conv_787)
%onnx::Conv_799 = Identity(%onnx::Conv_787)
%onnx::Conv_796 = Identity(%onnx::Conv_787)
%onnx::Conv_793 = Identity(%onnx::Conv_787)
%onnx::Conv_790 = Identity(%onnx::Conv_787)
%onnx::Conv_784 = Identity(%onnx::Conv_751)
%onnx::Conv_781 = Identity(%onnx::Conv_751)
%onnx::Conv_778 = Identity(%onnx::Conv_751)
%onnx::Conv_775 = Identity(%onnx::Conv_751)
%onnx::Conv_772 = Identity(%onnx::Conv_751)
%onnx::Conv_769 = Identity(%onnx::Conv_751)
%onnx::Conv_766 = Identity(%onnx::Conv_754)
%onnx::Conv_763 = Identity(%onnx::Conv_754)
%onnx::Conv_760 = Identity(%onnx::Conv_751)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_748 = Identity(%onnx::Conv_715)
%onnx::Conv_745 = Identity(%onnx::Conv_715)
%onnx::Conv_742 = Identity(%onnx::Conv_715)
%onnx::Conv_739 = Identity(%onnx::Conv_715)
%onnx::Conv_736 = Identity(%onnx::Conv_715)
%onnx::Conv_733 = Identity(%onnx::Conv_715)
%onnx::Conv_730 = Identity(%onnx::Conv_715)
%onnx::Conv_727 = Identity(%onnx::Conv_715)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_715)
%onnx::Conv_718 = Identity(%onnx::Conv_715)
%onnx::Conv_712 = Identity(%onnx::Conv_700)
%onnx::Conv_709 = Identity(%onnx::Conv_700)
%onnx::Conv_706 = Identity(%onnx::Conv_688)
%onnx::Conv_703 = Identity(%onnx::Conv_700)
%onnx::Conv_697 = Identity(%onnx::Conv_688)
%onnx::Conv_694 = Identity(%onnx::Conv_691)
%onnx::Conv_685 = Identity(%onnx::Conv_673)
%onnx::Conv_682 = Identity(%onnx::Conv_673)
%onnx::Conv_679 = Identity(%onnx::Conv_670)
%onnx::Conv_676 = Identity(%onnx::Conv_673)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_669, %onnx::Conv_670)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%667 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %667
}
|
val_accuracy
| 0 | 67,749,504 | 1,366,244 |
{'zcp_synflow': 79.09038295292683, 'zcp_zen': 67.83836364746094, 'zcp_epe_nas': 15.35390932225001, 'zcp_fisher': 0.1426919549703598, 'zcp_flops': 67749504.0, 'zcp_grad_norm': 28.142316818237305, 'zcp_grasp': 0.21132659912109375, 'zcp_jacov': -16.05399762779565, 'zcp_l2_norm': 582.9414672851562, 'zcp_nwot': 216.19027585721562, 'zcp_params': 1366244.0, 'zcp_plain': -0.0019004337955266237, 'zcp_snip': 43.987003326416016, 'lat_1080ti_1': 0.6143669420318413, 'lat_1080ti_32': 0.730611015154252, 'lat_1080ti_64': 0.7469771697073182, 'lat_2080ti_1': 0.653956194485593, 'lat_2080ti_32': 0.7463120601639703, 'lat_2080ti_64': 0.756064781223554, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.5150917792896046, 'lat_fpga': 0.3612741072047021, 'lat_gold_6226': 0.1990374351519924, 'lat_gold_6240': 0.33958311727432167, 'lat_pixel2': 0.391304347826087, 'lat_pixel3': 0.5478273656053008, 'lat_raspi4': 0.5031973634630189, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.1889763779527559, 'lat_silver_4114': 0.3494342675283171, 'lat_silver_4210r': 0.3518214911191891, 'lat_titan_rtx_1': 0.6258404774391534, 'lat_titan_rtx_32': 0.6881415872333718, 'lat_titan_rtx_64': 0.7803150421912524, 'lat_titanx_1': 0.32544651790624424, 'lat_titanx_32': 0.7600000121778566, 'lat_titanx_64': 0.7053729823042203, 'lat_titanxp_1': 0.5799453429892136, 'lat_titanxp_32': 0.7681765483089665, 'lat_titanxp_64': 0.7710293660631061}
| |
FBNet_2278
|
FBNet
|
2278
|
2278
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_515[FLOAT, 16x3x3x3]
%onnx::Conv_516[FLOAT, 16]
%onnx::Conv_518[FLOAT, 96x16x1x1]
%onnx::Conv_519[FLOAT, 96]
%onnx::Conv_521[FLOAT, 96x1x3x3]
%onnx::Conv_524[FLOAT, 16x96x1x1]
%onnx::Conv_527[FLOAT, 16x16x1x1]
%onnx::Conv_530[FLOAT, 16x1x3x3]
%onnx::Conv_533[FLOAT, 24x16x1x1]
%onnx::Conv_534[FLOAT, 24]
%onnx::Conv_536[FLOAT, 144x24x1x1]
%onnx::Conv_537[FLOAT, 144]
%onnx::Conv_539[FLOAT, 144x1x3x3]
%onnx::Conv_542[FLOAT, 24x144x1x1]
%onnx::Conv_545[FLOAT, 24x24x1x1]
%onnx::Conv_548[FLOAT, 24x1x3x3]
%onnx::Conv_551[FLOAT, 24x24x1x1]
%onnx::Conv_554[FLOAT, 72x24x1x1]
%onnx::Conv_555[FLOAT, 72]
%onnx::Conv_557[FLOAT, 72x1x5x5]
%onnx::Conv_560[FLOAT, 32x72x1x1]
%onnx::Conv_561[FLOAT, 32]
%onnx::Conv_563[FLOAT, 32x16x1x1]
%onnx::Conv_566[FLOAT, 32x1x5x5]
%onnx::Conv_569[FLOAT, 32x16x1x1]
%onnx::Conv_572[FLOAT, 192x32x1x1]
%onnx::Conv_573[FLOAT, 192]
%onnx::Conv_575[FLOAT, 192x1x3x3]
%onnx::Conv_578[FLOAT, 32x192x1x1]
%onnx::Conv_581[FLOAT, 32x32x1x1]
%onnx::Conv_584[FLOAT, 32x1x3x3]
%onnx::Conv_587[FLOAT, 32x32x1x1]
%onnx::Conv_590[FLOAT, 64x32x1x1]
%onnx::Conv_591[FLOAT, 64]
%onnx::Conv_593[FLOAT, 192x64x1x1]
%onnx::Conv_596[FLOAT, 192x1x3x3]
%onnx::Conv_599[FLOAT, 64x192x1x1]
%onnx::Conv_602[FLOAT, 64x32x1x1]
%onnx::Conv_605[FLOAT, 64x1x3x3]
%onnx::Conv_608[FLOAT, 64x32x1x1]
%onnx::Conv_611[FLOAT, 112x64x1x1]
%onnx::Conv_612[FLOAT, 112]
%onnx::Conv_614[FLOAT, 336x112x1x1]
%onnx::Conv_615[FLOAT, 336]
%onnx::Conv_617[FLOAT, 336x1x5x5]
%onnx::Conv_620[FLOAT, 112x336x1x1]
%onnx::Conv_623[FLOAT, 112x112x1x1]
%onnx::Conv_626[FLOAT, 112x1x3x3]
%onnx::Conv_629[FLOAT, 112x112x1x1]
%onnx::Conv_632[FLOAT, 336x112x1x1]
%onnx::Conv_635[FLOAT, 336x1x5x5]
%onnx::Conv_638[FLOAT, 184x336x1x1]
%onnx::Conv_639[FLOAT, 184]
%onnx::Conv_641[FLOAT, 552x184x1x1]
%onnx::Conv_642[FLOAT, 552]
%onnx::Conv_644[FLOAT, 552x1x5x5]
%onnx::Conv_647[FLOAT, 184x552x1x1]
%onnx::Conv_650[FLOAT, 1104x184x1x1]
%onnx::Conv_651[FLOAT, 1104]
%onnx::Conv_653[FLOAT, 1104x1x3x3]
%onnx::Conv_656[FLOAT, 184x1104x1x1]
%onnx::Conv_659[FLOAT, 552x184x1x1]
%onnx::Conv_662[FLOAT, 552x1x3x3]
%onnx::Conv_665[FLOAT, 352x552x1x1]
%onnx::Conv_666[FLOAT, 352]
%onnx::Conv_668[FLOAT, 1504x352x1x1]
%onnx::Conv_669[FLOAT, 1504]
) {
%onnx::Conv_663 = Identity(%onnx::Conv_642)
%onnx::Conv_660 = Identity(%onnx::Conv_642)
%onnx::Conv_657 = Identity(%onnx::Conv_639)
%onnx::Conv_654 = Identity(%onnx::Conv_651)
%onnx::Conv_648 = Identity(%onnx::Conv_639)
%onnx::Conv_645 = Identity(%onnx::Conv_642)
%onnx::Conv_636 = Identity(%onnx::Conv_615)
%onnx::Conv_633 = Identity(%onnx::Conv_615)
%onnx::Conv_630 = Identity(%onnx::Conv_612)
%onnx::Conv_627 = Identity(%onnx::Conv_612)
%onnx::Conv_624 = Identity(%onnx::Conv_612)
%onnx::Conv_621 = Identity(%onnx::Conv_612)
%onnx::Conv_618 = Identity(%onnx::Conv_615)
%onnx::Conv_609 = Identity(%onnx::Conv_591)
%onnx::Conv_606 = Identity(%onnx::Conv_591)
%onnx::Conv_603 = Identity(%onnx::Conv_591)
%onnx::Conv_600 = Identity(%onnx::Conv_591)
%onnx::Conv_597 = Identity(%onnx::Conv_573)
%onnx::Conv_594 = Identity(%onnx::Conv_573)
%onnx::Conv_588 = Identity(%onnx::Conv_561)
%onnx::Conv_585 = Identity(%onnx::Conv_561)
%onnx::Conv_582 = Identity(%onnx::Conv_561)
%onnx::Conv_579 = Identity(%onnx::Conv_561)
%onnx::Conv_576 = Identity(%onnx::Conv_573)
%onnx::Conv_570 = Identity(%onnx::Conv_561)
%onnx::Conv_567 = Identity(%onnx::Conv_561)
%onnx::Conv_564 = Identity(%onnx::Conv_561)
%onnx::Conv_558 = Identity(%onnx::Conv_555)
%onnx::Conv_552 = Identity(%onnx::Conv_534)
%onnx::Conv_549 = Identity(%onnx::Conv_534)
%onnx::Conv_546 = Identity(%onnx::Conv_534)
%onnx::Conv_543 = Identity(%onnx::Conv_534)
%onnx::Conv_540 = Identity(%onnx::Conv_537)
%onnx::Conv_531 = Identity(%onnx::Conv_516)
%onnx::Conv_528 = Identity(%onnx::Conv_516)
%onnx::Conv_525 = Identity(%onnx::Conv_516)
%onnx::Conv_522 = Identity(%onnx::Conv_519)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_515, %onnx::Conv_516)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_518, %onnx::Conv_519)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_521, %onnx::Conv_522)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_524, %onnx::Conv_525)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_527, %onnx::Conv_528)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_530, %onnx::Conv_531)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_533, %onnx::Conv_534)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_536, %onnx::Conv_537)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_539, %onnx::Conv_540)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_542, %onnx::Conv_543)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_545, %onnx::Conv_546)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_548, %onnx::Conv_549)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_551, %onnx::Conv_552)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_554, %onnx::Conv_555)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_557, %onnx::Conv_558)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_560, %onnx::Conv_561)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_563, %onnx::Conv_564)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_566, %onnx::Conv_567)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_569, %onnx::Conv_570)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_572, %onnx::Conv_573)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_575, %onnx::Conv_576)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_578, %onnx::Conv_579)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_581, %onnx::Conv_582)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_584, %onnx::Conv_585)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_587, %onnx::Conv_588)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_590, %onnx::Conv_591)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_593, %onnx::Conv_594)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_596, %onnx::Conv_597)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_599, %onnx::Conv_600)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_602, %onnx::Conv_603)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_605, %onnx::Conv_606)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%513 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %513
}
|
val_accuracy
| 0 | 62,172,288 | 1,932,196 |
{'zcp_synflow': 67.41264867382752, 'zcp_zen': 56.80339050292969, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.06079770624637604, 'zcp_flops': 62172288.0, 'zcp_grad_norm': 17.4497127532959, 'zcp_grasp': -0.10856246948242188, 'zcp_jacov': -16.06829379932628, 'zcp_l2_norm': 542.722900390625, 'zcp_nwot': 212.24740822996566, 'zcp_params': 1932196.0, 'zcp_plain': -0.003140533110126853, 'zcp_snip': 29.92837905883789, 'lat_1080ti_1': 0.20424554553327928, 'lat_1080ti_32': 0.292535812755568, 'lat_1080ti_64': 0.26375521112700684, 'lat_2080ti_1': 0.23066059929849786, 'lat_2080ti_32': 0.25110660312224226, 'lat_2080ti_64': 0.28705555391530935, 'lat_essential_ph_1': 0.33962264150943394, 'lat_eyeriss': 0.35774668549318767, 'lat_fpga': 0.39844051494671423, 'lat_gold_6226': 0.3099981133590486, 'lat_gold_6240': 0.28554281113519203, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.3122628671304946, 'lat_raspi4': 0.3991724280900862, 'lat_samsung_a50': 0.15789473684210525, 'lat_samsung_s7': 0.11811023622047244, 'lat_silver_4114': 0.26825729244688273, 'lat_silver_4210r': 0.2487325760118517, 'lat_titan_rtx_1': 0.22323122725909397, 'lat_titan_rtx_32': 0.2245809259658103, 'lat_titan_rtx_64': 0.26414045097706756, 'lat_titanx_1': 0.11846644137097252, 'lat_titanx_32': 0.22344395644639328, 'lat_titanx_64': 0.25774968681874405, 'lat_titanxp_1': 0.2064082077621074, 'lat_titanxp_32': 0.23215377798156447, 'lat_titanxp_64': 0.272178717719814}
| |
FBNet_4599
|
FBNet
|
4599
|
4599
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_605[FLOAT, 16x3x3x3]
%onnx::Conv_606[FLOAT, 16]
%onnx::Conv_608[FLOAT, 48x16x1x1]
%onnx::Conv_609[FLOAT, 48]
%onnx::Conv_611[FLOAT, 48x1x5x5]
%onnx::Conv_614[FLOAT, 16x48x1x1]
%onnx::Conv_617[FLOAT, 16x8x1x1]
%onnx::Conv_620[FLOAT, 16x1x3x3]
%onnx::Conv_623[FLOAT, 24x8x1x1]
%onnx::Conv_624[FLOAT, 24]
%onnx::Conv_626[FLOAT, 72x24x1x1]
%onnx::Conv_627[FLOAT, 72]
%onnx::Conv_629[FLOAT, 72x1x3x3]
%onnx::Conv_632[FLOAT, 24x72x1x1]
%onnx::Conv_635[FLOAT, 144x24x1x1]
%onnx::Conv_636[FLOAT, 144]
%onnx::Conv_638[FLOAT, 144x1x5x5]
%onnx::Conv_641[FLOAT, 24x144x1x1]
%onnx::Conv_644[FLOAT, 24x12x1x1]
%onnx::Conv_647[FLOAT, 24x1x5x5]
%onnx::Conv_650[FLOAT, 24x12x1x1]
%onnx::Conv_653[FLOAT, 144x24x1x1]
%onnx::Conv_656[FLOAT, 144x1x5x5]
%onnx::Conv_659[FLOAT, 32x144x1x1]
%onnx::Conv_660[FLOAT, 32]
%onnx::Conv_662[FLOAT, 32x16x1x1]
%onnx::Conv_665[FLOAT, 32x1x5x5]
%onnx::Conv_668[FLOAT, 32x16x1x1]
%onnx::Conv_671[FLOAT, 32x32x1x1]
%onnx::Conv_674[FLOAT, 32x1x3x3]
%onnx::Conv_677[FLOAT, 32x32x1x1]
%onnx::Conv_680[FLOAT, 192x32x1x1]
%onnx::Conv_681[FLOAT, 192]
%onnx::Conv_683[FLOAT, 192x1x3x3]
%onnx::Conv_686[FLOAT, 64x192x1x1]
%onnx::Conv_687[FLOAT, 64]
%onnx::Conv_689[FLOAT, 192x64x1x1]
%onnx::Conv_692[FLOAT, 192x1x3x3]
%onnx::Conv_695[FLOAT, 64x192x1x1]
%onnx::Conv_698[FLOAT, 192x64x1x1]
%onnx::Conv_701[FLOAT, 192x1x3x3]
%onnx::Conv_704[FLOAT, 64x192x1x1]
%onnx::Conv_707[FLOAT, 192x64x1x1]
%onnx::Conv_710[FLOAT, 192x1x3x3]
%onnx::Conv_713[FLOAT, 64x192x1x1]
%onnx::Conv_716[FLOAT, 192x64x1x1]
%onnx::Conv_719[FLOAT, 192x1x3x3]
%onnx::Conv_722[FLOAT, 112x192x1x1]
%onnx::Conv_723[FLOAT, 112]
%onnx::Conv_725[FLOAT, 672x112x1x1]
%onnx::Conv_726[FLOAT, 672]
%onnx::Conv_728[FLOAT, 672x1x3x3]
%onnx::Conv_731[FLOAT, 112x672x1x1]
%onnx::Conv_734[FLOAT, 336x112x1x1]
%onnx::Conv_735[FLOAT, 336]
%onnx::Conv_737[FLOAT, 336x1x5x5]
%onnx::Conv_740[FLOAT, 112x336x1x1]
%onnx::Conv_743[FLOAT, 184x112x1x1]
%onnx::Conv_744[FLOAT, 184]
%onnx::Conv_746[FLOAT, 184x184x1x1]
%onnx::Conv_749[FLOAT, 184x1x3x3]
%onnx::Conv_752[FLOAT, 184x184x1x1]
%onnx::Conv_755[FLOAT, 184x184x1x1]
%onnx::Conv_758[FLOAT, 184x1x5x5]
%onnx::Conv_761[FLOAT, 184x184x1x1]
%onnx::Conv_764[FLOAT, 184x184x1x1]
%onnx::Conv_767[FLOAT, 184x1x5x5]
%onnx::Conv_770[FLOAT, 184x184x1x1]
%onnx::Conv_773[FLOAT, 1104x184x1x1]
%onnx::Conv_774[FLOAT, 1104]
%onnx::Conv_776[FLOAT, 1104x1x3x3]
%onnx::Conv_779[FLOAT, 352x1104x1x1]
%onnx::Conv_780[FLOAT, 352]
%onnx::Conv_782[FLOAT, 1504x352x1x1]
%onnx::Conv_783[FLOAT, 1504]
) {
%onnx::Conv_777 = Identity(%onnx::Conv_774)
%onnx::Conv_771 = Identity(%onnx::Conv_744)
%onnx::Conv_768 = Identity(%onnx::Conv_744)
%onnx::Conv_765 = Identity(%onnx::Conv_744)
%onnx::Conv_762 = Identity(%onnx::Conv_744)
%onnx::Conv_759 = Identity(%onnx::Conv_744)
%onnx::Conv_756 = Identity(%onnx::Conv_744)
%onnx::Conv_753 = Identity(%onnx::Conv_744)
%onnx::Conv_750 = Identity(%onnx::Conv_744)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_741 = Identity(%onnx::Conv_723)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_732 = Identity(%onnx::Conv_723)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_720 = Identity(%onnx::Conv_681)
%onnx::Conv_717 = Identity(%onnx::Conv_681)
%onnx::Conv_714 = Identity(%onnx::Conv_687)
%onnx::Conv_711 = Identity(%onnx::Conv_681)
%onnx::Conv_708 = Identity(%onnx::Conv_681)
%onnx::Conv_705 = Identity(%onnx::Conv_687)
%onnx::Conv_702 = Identity(%onnx::Conv_681)
%onnx::Conv_699 = Identity(%onnx::Conv_681)
%onnx::Conv_696 = Identity(%onnx::Conv_687)
%onnx::Conv_693 = Identity(%onnx::Conv_681)
%onnx::Conv_690 = Identity(%onnx::Conv_681)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_678 = Identity(%onnx::Conv_660)
%onnx::Conv_675 = Identity(%onnx::Conv_660)
%onnx::Conv_672 = Identity(%onnx::Conv_660)
%onnx::Conv_669 = Identity(%onnx::Conv_660)
%onnx::Conv_666 = Identity(%onnx::Conv_660)
%onnx::Conv_663 = Identity(%onnx::Conv_660)
%onnx::Conv_657 = Identity(%onnx::Conv_636)
%onnx::Conv_654 = Identity(%onnx::Conv_636)
%onnx::Conv_651 = Identity(%onnx::Conv_624)
%onnx::Conv_648 = Identity(%onnx::Conv_624)
%onnx::Conv_645 = Identity(%onnx::Conv_624)
%onnx::Conv_642 = Identity(%onnx::Conv_624)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%onnx::Conv_633 = Identity(%onnx::Conv_624)
%onnx::Conv_630 = Identity(%onnx::Conv_627)
%onnx::Conv_621 = Identity(%onnx::Conv_606)
%onnx::Conv_618 = Identity(%onnx::Conv_606)
%onnx::Conv_615 = Identity(%onnx::Conv_606)
%onnx::Conv_612 = Identity(%onnx::Conv_609)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_605, %onnx::Conv_606)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%603 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %603
}
|
val_accuracy
| 0 | 77,182,976 | 1,949,628 |
{'zcp_synflow': 76.194169453985, 'zcp_zen': 66.50231170654297, 'zcp_epe_nas': 22.895714173450358, 'zcp_fisher': 0.12591169774532318, 'zcp_flops': 77182976.0, 'zcp_grad_norm': 26.169769287109375, 'zcp_grasp': 0.06266975402832031, 'zcp_jacov': -16.053002824276092, 'zcp_l2_norm': 622.452392578125, 'zcp_nwot': 214.97640912496385, 'zcp_params': 1949628.0, 'zcp_plain': 0.002621118212118745, 'zcp_snip': 42.446163177490234, 'lat_1080ti_1': 0.51065111806754, 'lat_1080ti_32': 0.5600921156408543, 'lat_1080ti_64': 0.5354659521437186, 'lat_2080ti_1': 0.5509927071296734, 'lat_2080ti_32': 0.5416504052702189, 'lat_2080ti_64': 0.5480357211220916, 'lat_essential_ph_1': 0.33962264150943394, 'lat_eyeriss': 0.52314195074542, 'lat_fpga': 0.5550915801660415, 'lat_gold_6226': 0.365951388513599, 'lat_gold_6240': 0.43018087529680155, 'lat_pixel2': 0.32608695652173914, 'lat_pixel3': 0.560039044310902, 'lat_raspi4': 0.6066884779551199, 'lat_samsung_a50': 0.21052631578947367, 'lat_samsung_s7': 0.1732283464566929, 'lat_silver_4114': 0.45282620291243236, 'lat_silver_4210r': 0.526744037761711, 'lat_titan_rtx_1': 0.49104432396735376, 'lat_titan_rtx_32': 0.5889447943433593, 'lat_titan_rtx_64': 0.5377483361951856, 'lat_titanx_1': 0.260556482881934, 'lat_titanx_32': 0.5149914357909169, 'lat_titanx_64': 0.5661658472905927, 'lat_titanxp_1': 0.4921826588533068, 'lat_titanxp_32': 0.5362458378488747, 'lat_titanxp_64': 0.5415991590942653}
| |
FBNet_4855
|
FBNet
|
4855
|
4855
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_687[FLOAT, 16x3x3x3]
%onnx::Conv_688[FLOAT, 16]
%onnx::Conv_690[FLOAT, 48x16x1x1]
%onnx::Conv_691[FLOAT, 48]
%onnx::Conv_693[FLOAT, 48x1x5x5]
%onnx::Conv_696[FLOAT, 16x48x1x1]
%onnx::Conv_699[FLOAT, 48x16x1x1]
%onnx::Conv_702[FLOAT, 48x1x5x5]
%onnx::Conv_705[FLOAT, 24x48x1x1]
%onnx::Conv_706[FLOAT, 24]
%onnx::Conv_708[FLOAT, 72x24x1x1]
%onnx::Conv_709[FLOAT, 72]
%onnx::Conv_711[FLOAT, 72x1x3x3]
%onnx::Conv_714[FLOAT, 24x72x1x1]
%onnx::Conv_717[FLOAT, 24x12x1x1]
%onnx::Conv_720[FLOAT, 24x1x5x5]
%onnx::Conv_723[FLOAT, 24x12x1x1]
%onnx::Conv_726[FLOAT, 72x24x1x1]
%onnx::Conv_729[FLOAT, 72x1x5x5]
%onnx::Conv_732[FLOAT, 32x72x1x1]
%onnx::Conv_733[FLOAT, 32]
%onnx::Conv_735[FLOAT, 32x16x1x1]
%onnx::Conv_738[FLOAT, 32x1x3x3]
%onnx::Conv_741[FLOAT, 32x16x1x1]
%onnx::Conv_744[FLOAT, 32x32x1x1]
%onnx::Conv_747[FLOAT, 32x1x3x3]
%onnx::Conv_750[FLOAT, 32x32x1x1]
%onnx::Conv_753[FLOAT, 32x16x1x1]
%onnx::Conv_756[FLOAT, 32x1x5x5]
%onnx::Conv_759[FLOAT, 32x16x1x1]
%onnx::Conv_762[FLOAT, 192x32x1x1]
%onnx::Conv_763[FLOAT, 192]
%onnx::Conv_765[FLOAT, 192x1x3x3]
%onnx::Conv_768[FLOAT, 64x192x1x1]
%onnx::Conv_769[FLOAT, 64]
%onnx::Conv_771[FLOAT, 384x64x1x1]
%onnx::Conv_772[FLOAT, 384]
%onnx::Conv_774[FLOAT, 384x1x3x3]
%onnx::Conv_777[FLOAT, 64x384x1x1]
%onnx::Conv_780[FLOAT, 64x32x1x1]
%onnx::Conv_783[FLOAT, 64x1x5x5]
%onnx::Conv_786[FLOAT, 64x32x1x1]
%onnx::Conv_789[FLOAT, 384x64x1x1]
%onnx::Conv_792[FLOAT, 384x1x3x3]
%onnx::Conv_795[FLOAT, 112x384x1x1]
%onnx::Conv_796[FLOAT, 112]
%onnx::Conv_798[FLOAT, 112x112x1x1]
%onnx::Conv_801[FLOAT, 112x1x3x3]
%onnx::Conv_804[FLOAT, 112x112x1x1]
%onnx::Conv_807[FLOAT, 112x56x1x1]
%onnx::Conv_810[FLOAT, 112x1x3x3]
%onnx::Conv_813[FLOAT, 112x56x1x1]
%onnx::Conv_816[FLOAT, 112x56x1x1]
%onnx::Conv_819[FLOAT, 112x1x3x3]
%onnx::Conv_822[FLOAT, 184x56x1x1]
%onnx::Conv_823[FLOAT, 184]
%onnx::Conv_825[FLOAT, 1104x184x1x1]
%onnx::Conv_826[FLOAT, 1104]
%onnx::Conv_828[FLOAT, 1104x1x5x5]
%onnx::Conv_831[FLOAT, 184x1104x1x1]
%onnx::Conv_834[FLOAT, 552x184x1x1]
%onnx::Conv_835[FLOAT, 552]
%onnx::Conv_837[FLOAT, 552x1x3x3]
%onnx::Conv_840[FLOAT, 184x552x1x1]
%onnx::Conv_843[FLOAT, 184x92x1x1]
%onnx::Conv_846[FLOAT, 184x1x5x5]
%onnx::Conv_849[FLOAT, 184x92x1x1]
%onnx::Conv_852[FLOAT, 184x92x1x1]
%onnx::Conv_855[FLOAT, 184x1x5x5]
%onnx::Conv_858[FLOAT, 352x92x1x1]
%onnx::Conv_859[FLOAT, 352]
%onnx::Conv_861[FLOAT, 1504x352x1x1]
%onnx::Conv_862[FLOAT, 1504]
) {
%onnx::Conv_856 = Identity(%onnx::Conv_823)
%onnx::Conv_853 = Identity(%onnx::Conv_823)
%onnx::Conv_850 = Identity(%onnx::Conv_823)
%onnx::Conv_847 = Identity(%onnx::Conv_823)
%onnx::Conv_844 = Identity(%onnx::Conv_823)
%onnx::Conv_841 = Identity(%onnx::Conv_823)
%onnx::Conv_838 = Identity(%onnx::Conv_835)
%onnx::Conv_832 = Identity(%onnx::Conv_823)
%onnx::Conv_829 = Identity(%onnx::Conv_826)
%onnx::Conv_820 = Identity(%onnx::Conv_796)
%onnx::Conv_817 = Identity(%onnx::Conv_796)
%onnx::Conv_814 = Identity(%onnx::Conv_796)
%onnx::Conv_811 = Identity(%onnx::Conv_796)
%onnx::Conv_808 = Identity(%onnx::Conv_796)
%onnx::Conv_805 = Identity(%onnx::Conv_796)
%onnx::Conv_802 = Identity(%onnx::Conv_796)
%onnx::Conv_799 = Identity(%onnx::Conv_796)
%onnx::Conv_793 = Identity(%onnx::Conv_772)
%onnx::Conv_790 = Identity(%onnx::Conv_772)
%onnx::Conv_787 = Identity(%onnx::Conv_769)
%onnx::Conv_784 = Identity(%onnx::Conv_769)
%onnx::Conv_781 = Identity(%onnx::Conv_769)
%onnx::Conv_778 = Identity(%onnx::Conv_769)
%onnx::Conv_775 = Identity(%onnx::Conv_772)
%onnx::Conv_766 = Identity(%onnx::Conv_763)
%onnx::Conv_760 = Identity(%onnx::Conv_733)
%onnx::Conv_757 = Identity(%onnx::Conv_733)
%onnx::Conv_754 = Identity(%onnx::Conv_733)
%onnx::Conv_751 = Identity(%onnx::Conv_733)
%onnx::Conv_748 = Identity(%onnx::Conv_733)
%onnx::Conv_745 = Identity(%onnx::Conv_733)
%onnx::Conv_742 = Identity(%onnx::Conv_733)
%onnx::Conv_739 = Identity(%onnx::Conv_733)
%onnx::Conv_736 = Identity(%onnx::Conv_733)
%onnx::Conv_730 = Identity(%onnx::Conv_709)
%onnx::Conv_727 = Identity(%onnx::Conv_709)
%onnx::Conv_724 = Identity(%onnx::Conv_706)
%onnx::Conv_721 = Identity(%onnx::Conv_706)
%onnx::Conv_718 = Identity(%onnx::Conv_706)
%onnx::Conv_715 = Identity(%onnx::Conv_706)
%onnx::Conv_712 = Identity(%onnx::Conv_709)
%onnx::Conv_703 = Identity(%onnx::Conv_691)
%onnx::Conv_700 = Identity(%onnx::Conv_691)
%onnx::Conv_697 = Identity(%onnx::Conv_688)
%onnx::Conv_694 = Identity(%onnx::Conv_691)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_687, %onnx::Conv_688)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%685 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %685
}
|
val_accuracy
| 0 | 52,761,728 | 1,665,540 |
{'zcp_synflow': 67.47701333090478, 'zcp_zen': 60.84530258178711, 'zcp_epe_nas': 7.4736721408808116, 'zcp_fisher': 0.11787315458059311, 'zcp_flops': 52761728.0, 'zcp_grad_norm': 24.177303314208984, 'zcp_grasp': -0.02095794677734375, 'zcp_jacov': -16.09285314854519, 'zcp_l2_norm': 542.2675170898438, 'zcp_nwot': 208.69347948297715, 'zcp_params': 1665540.0, 'zcp_plain': 0.00021975085837766528, 'zcp_snip': 38.403717041015625, 'lat_1080ti_1': 0.5324143504397403, 'lat_1080ti_32': 0.5210202363168197, 'lat_1080ti_64': 0.3299930169488742, 'lat_2080ti_1': 0.6192643119378379, 'lat_2080ti_32': 0.45605279844194047, 'lat_2080ti_64': 0.3506600914456223, 'lat_essential_ph_1': 0.1320754716981132, 'lat_eyeriss': 0.28312214622338766, 'lat_fpga': 0.2638161865840748, 'lat_gold_6226': 0.26534370045808975, 'lat_gold_6240': 0.35506092399441136, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.27462029610287914, 'lat_raspi4': 0.2813988689410126, 'lat_samsung_a50': 0.1368421052631579, 'lat_samsung_s7': 0.13385826771653545, 'lat_silver_4114': 0.4078654483965893, 'lat_silver_4210r': 0.4569113737748346, 'lat_titan_rtx_1': 0.5545629585065525, 'lat_titan_rtx_32': 0.4533861651009249, 'lat_titan_rtx_64': 0.3680615763863577, 'lat_titanx_1': 0.28733009923434827, 'lat_titanx_32': 0.3951551449530437, 'lat_titanx_64': 0.3155040180438069, 'lat_titanxp_1': 0.5414979296816478, 'lat_titanxp_32': 0.4214928863145757, 'lat_titanxp_64': 0.35517032953547534}
| |
FBNet_2435
|
FBNet
|
2435
|
2435
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_575[FLOAT, 16x3x3x3]
%onnx::Conv_576[FLOAT, 16]
%onnx::Conv_578[FLOAT, 48x16x1x1]
%onnx::Conv_579[FLOAT, 48]
%onnx::Conv_581[FLOAT, 48x1x5x5]
%onnx::Conv_584[FLOAT, 16x48x1x1]
%onnx::Conv_587[FLOAT, 16x16x1x1]
%onnx::Conv_590[FLOAT, 16x1x5x5]
%onnx::Conv_593[FLOAT, 24x16x1x1]
%onnx::Conv_594[FLOAT, 24]
%onnx::Conv_596[FLOAT, 24x12x1x1]
%onnx::Conv_599[FLOAT, 24x1x5x5]
%onnx::Conv_602[FLOAT, 24x12x1x1]
%onnx::Conv_605[FLOAT, 24x24x1x1]
%onnx::Conv_608[FLOAT, 24x1x5x5]
%onnx::Conv_611[FLOAT, 24x24x1x1]
%onnx::Conv_614[FLOAT, 24x12x1x1]
%onnx::Conv_617[FLOAT, 24x1x3x3]
%onnx::Conv_620[FLOAT, 32x12x1x1]
%onnx::Conv_621[FLOAT, 32]
%onnx::Conv_623[FLOAT, 192x32x1x1]
%onnx::Conv_624[FLOAT, 192]
%onnx::Conv_626[FLOAT, 192x1x3x3]
%onnx::Conv_629[FLOAT, 32x192x1x1]
%onnx::Conv_632[FLOAT, 32x32x1x1]
%onnx::Conv_635[FLOAT, 32x1x5x5]
%onnx::Conv_638[FLOAT, 32x32x1x1]
%onnx::Conv_641[FLOAT, 32x32x1x1]
%onnx::Conv_644[FLOAT, 32x1x3x3]
%onnx::Conv_647[FLOAT, 32x32x1x1]
%onnx::Conv_650[FLOAT, 192x32x1x1]
%onnx::Conv_653[FLOAT, 192x1x5x5]
%onnx::Conv_656[FLOAT, 64x192x1x1]
%onnx::Conv_657[FLOAT, 64]
%onnx::Conv_659[FLOAT, 384x64x1x1]
%onnx::Conv_660[FLOAT, 384]
%onnx::Conv_662[FLOAT, 384x1x3x3]
%onnx::Conv_665[FLOAT, 64x384x1x1]
%onnx::Conv_668[FLOAT, 384x64x1x1]
%onnx::Conv_671[FLOAT, 384x1x3x3]
%onnx::Conv_674[FLOAT, 64x384x1x1]
%onnx::Conv_677[FLOAT, 192x64x1x1]
%onnx::Conv_680[FLOAT, 192x1x5x5]
%onnx::Conv_683[FLOAT, 64x192x1x1]
%onnx::Conv_686[FLOAT, 64x64x1x1]
%onnx::Conv_689[FLOAT, 64x1x3x3]
%onnx::Conv_692[FLOAT, 112x64x1x1]
%onnx::Conv_693[FLOAT, 112]
%onnx::Conv_695[FLOAT, 672x112x1x1]
%onnx::Conv_696[FLOAT, 672]
%onnx::Conv_698[FLOAT, 672x1x3x3]
%onnx::Conv_701[FLOAT, 112x672x1x1]
%onnx::Conv_704[FLOAT, 672x112x1x1]
%onnx::Conv_707[FLOAT, 672x1x3x3]
%onnx::Conv_710[FLOAT, 112x672x1x1]
%onnx::Conv_713[FLOAT, 112x112x1x1]
%onnx::Conv_716[FLOAT, 112x1x3x3]
%onnx::Conv_719[FLOAT, 184x112x1x1]
%onnx::Conv_720[FLOAT, 184]
%onnx::Conv_722[FLOAT, 552x184x1x1]
%onnx::Conv_723[FLOAT, 552]
%onnx::Conv_725[FLOAT, 552x1x3x3]
%onnx::Conv_728[FLOAT, 184x552x1x1]
%onnx::Conv_731[FLOAT, 184x184x1x1]
%onnx::Conv_734[FLOAT, 184x1x3x3]
%onnx::Conv_737[FLOAT, 184x184x1x1]
%onnx::Conv_740[FLOAT, 1104x184x1x1]
%onnx::Conv_741[FLOAT, 1104]
%onnx::Conv_743[FLOAT, 1104x1x3x3]
%onnx::Conv_746[FLOAT, 352x1104x1x1]
%onnx::Conv_747[FLOAT, 352]
%onnx::Conv_749[FLOAT, 1504x352x1x1]
%onnx::Conv_750[FLOAT, 1504]
) {
%onnx::Conv_744 = Identity(%onnx::Conv_741)
%onnx::Conv_738 = Identity(%onnx::Conv_720)
%onnx::Conv_735 = Identity(%onnx::Conv_720)
%onnx::Conv_732 = Identity(%onnx::Conv_720)
%onnx::Conv_729 = Identity(%onnx::Conv_720)
%onnx::Conv_726 = Identity(%onnx::Conv_723)
%onnx::Conv_717 = Identity(%onnx::Conv_693)
%onnx::Conv_714 = Identity(%onnx::Conv_693)
%onnx::Conv_711 = Identity(%onnx::Conv_693)
%onnx::Conv_708 = Identity(%onnx::Conv_696)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_693)
%onnx::Conv_699 = Identity(%onnx::Conv_696)
%onnx::Conv_690 = Identity(%onnx::Conv_657)
%onnx::Conv_687 = Identity(%onnx::Conv_657)
%onnx::Conv_684 = Identity(%onnx::Conv_657)
%onnx::Conv_681 = Identity(%onnx::Conv_624)
%onnx::Conv_678 = Identity(%onnx::Conv_624)
%onnx::Conv_675 = Identity(%onnx::Conv_657)
%onnx::Conv_672 = Identity(%onnx::Conv_660)
%onnx::Conv_669 = Identity(%onnx::Conv_660)
%onnx::Conv_666 = Identity(%onnx::Conv_657)
%onnx::Conv_663 = Identity(%onnx::Conv_660)
%onnx::Conv_654 = Identity(%onnx::Conv_624)
%onnx::Conv_651 = Identity(%onnx::Conv_624)
%onnx::Conv_648 = Identity(%onnx::Conv_621)
%onnx::Conv_645 = Identity(%onnx::Conv_621)
%onnx::Conv_642 = Identity(%onnx::Conv_621)
%onnx::Conv_639 = Identity(%onnx::Conv_621)
%onnx::Conv_636 = Identity(%onnx::Conv_621)
%onnx::Conv_633 = Identity(%onnx::Conv_621)
%onnx::Conv_630 = Identity(%onnx::Conv_621)
%onnx::Conv_627 = Identity(%onnx::Conv_624)
%onnx::Conv_618 = Identity(%onnx::Conv_594)
%onnx::Conv_615 = Identity(%onnx::Conv_594)
%onnx::Conv_612 = Identity(%onnx::Conv_594)
%onnx::Conv_609 = Identity(%onnx::Conv_594)
%onnx::Conv_606 = Identity(%onnx::Conv_594)
%onnx::Conv_603 = Identity(%onnx::Conv_594)
%onnx::Conv_600 = Identity(%onnx::Conv_594)
%onnx::Conv_597 = Identity(%onnx::Conv_594)
%onnx::Conv_591 = Identity(%onnx::Conv_576)
%onnx::Conv_588 = Identity(%onnx::Conv_576)
%onnx::Conv_585 = Identity(%onnx::Conv_576)
%onnx::Conv_582 = Identity(%onnx::Conv_579)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_575, %onnx::Conv_576)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_578, %onnx::Conv_579)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_581, %onnx::Conv_582)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_584, %onnx::Conv_585)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_587, %onnx::Conv_588)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_590, %onnx::Conv_591)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_593, %onnx::Conv_594)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_596, %onnx::Conv_597)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_599, %onnx::Conv_600)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_602, %onnx::Conv_603)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_605, %onnx::Conv_606)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%573 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %573
}
|
val_accuracy
| 0 | 71,111,296 | 2,129,340 |
{'zcp_synflow': 75.4167033519304, 'zcp_zen': 65.40941619873047, 'zcp_epe_nas': 7.505013258397767, 'zcp_fisher': 0.08659812062978745, 'zcp_flops': 71111296.0, 'zcp_grad_norm': 23.91156005859375, 'zcp_grasp': -0.053142547607421875, 'zcp_jacov': -16.07164323886066, 'zcp_l2_norm': 626.0662841796875, 'zcp_nwot': 209.03221409418762, 'zcp_params': 2129340.0, 'zcp_plain': 0.003367864992469549, 'zcp_snip': 40.256813049316406, 'lat_1080ti_1': 0.41985520612976407, 'lat_1080ti_32': 0.30700642427814484, 'lat_1080ti_64': 0.21779324908476774, 'lat_2080ti_1': 0.4460215850500279, 'lat_2080ti_32': 0.323689490478573, 'lat_2080ti_64': 0.24812041565997964, 'lat_essential_ph_1': 0.16981132075471697, 'lat_eyeriss': 0.42548548585260315, 'lat_fpga': 0.5467301279450616, 'lat_gold_6226': 0.4254306426930667, 'lat_gold_6240': 0.5785762596040006, 'lat_pixel2': 0.3695652173913043, 'lat_pixel3': 0.40502451206367973, 'lat_raspi4': 0.4663737946245125, 'lat_samsung_a50': 0.2, 'lat_samsung_s7': 0.1732283464566929, 'lat_silver_4114': 0.5533014503737097, 'lat_silver_4210r': 0.5825154262436728, 'lat_titan_rtx_1': 0.42911009391436133, 'lat_titan_rtx_32': 0.32986775859842554, 'lat_titan_rtx_64': 0.23625594741383796, 'lat_titanx_1': 0.24601759005648935, 'lat_titanx_32': 0.26877051603403773, 'lat_titanx_64': 0.22606911757412643, 'lat_titanxp_1': 0.42100968196029315, 'lat_titanxp_32': 0.2996068609903505, 'lat_titanxp_64': 0.23750202688657143}
| |
FBNet_4105
|
FBNet
|
4105
|
4105
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_713[FLOAT, 16x3x3x3]
%onnx::Conv_714[FLOAT, 16]
%onnx::Conv_716[FLOAT, 16x16x1x1]
%onnx::Conv_719[FLOAT, 16x1x5x5]
%onnx::Conv_722[FLOAT, 16x16x1x1]
%onnx::Conv_725[FLOAT, 48x16x1x1]
%onnx::Conv_726[FLOAT, 48]
%onnx::Conv_728[FLOAT, 48x1x3x3]
%onnx::Conv_731[FLOAT, 24x48x1x1]
%onnx::Conv_732[FLOAT, 24]
%onnx::Conv_734[FLOAT, 24x24x1x1]
%onnx::Conv_737[FLOAT, 24x1x5x5]
%onnx::Conv_740[FLOAT, 24x24x1x1]
%onnx::Conv_743[FLOAT, 72x24x1x1]
%onnx::Conv_744[FLOAT, 72]
%onnx::Conv_746[FLOAT, 72x1x5x5]
%onnx::Conv_749[FLOAT, 24x72x1x1]
%onnx::Conv_752[FLOAT, 144x24x1x1]
%onnx::Conv_753[FLOAT, 144]
%onnx::Conv_755[FLOAT, 144x1x5x5]
%onnx::Conv_758[FLOAT, 24x144x1x1]
%onnx::Conv_761[FLOAT, 72x24x1x1]
%onnx::Conv_764[FLOAT, 72x1x3x3]
%onnx::Conv_767[FLOAT, 32x72x1x1]
%onnx::Conv_768[FLOAT, 32]
%onnx::Conv_770[FLOAT, 32x16x1x1]
%onnx::Conv_773[FLOAT, 32x1x5x5]
%onnx::Conv_776[FLOAT, 32x16x1x1]
%onnx::Conv_779[FLOAT, 32x32x1x1]
%onnx::Conv_782[FLOAT, 32x1x5x5]
%onnx::Conv_785[FLOAT, 32x32x1x1]
%onnx::Conv_788[FLOAT, 32x16x1x1]
%onnx::Conv_791[FLOAT, 32x1x5x5]
%onnx::Conv_794[FLOAT, 32x16x1x1]
%onnx::Conv_797[FLOAT, 32x32x1x1]
%onnx::Conv_800[FLOAT, 32x1x5x5]
%onnx::Conv_803[FLOAT, 64x32x1x1]
%onnx::Conv_804[FLOAT, 64]
%onnx::Conv_806[FLOAT, 384x64x1x1]
%onnx::Conv_807[FLOAT, 384]
%onnx::Conv_809[FLOAT, 384x1x3x3]
%onnx::Conv_812[FLOAT, 64x384x1x1]
%onnx::Conv_815[FLOAT, 64x32x1x1]
%onnx::Conv_818[FLOAT, 64x1x5x5]
%onnx::Conv_821[FLOAT, 64x32x1x1]
%onnx::Conv_824[FLOAT, 192x64x1x1]
%onnx::Conv_825[FLOAT, 192]
%onnx::Conv_827[FLOAT, 192x1x3x3]
%onnx::Conv_830[FLOAT, 64x192x1x1]
%onnx::Conv_833[FLOAT, 384x64x1x1]
%onnx::Conv_836[FLOAT, 384x1x5x5]
%onnx::Conv_839[FLOAT, 112x384x1x1]
%onnx::Conv_840[FLOAT, 112]
%onnx::Conv_842[FLOAT, 672x112x1x1]
%onnx::Conv_843[FLOAT, 672]
%onnx::Conv_845[FLOAT, 672x1x5x5]
%onnx::Conv_848[FLOAT, 112x672x1x1]
%onnx::Conv_851[FLOAT, 336x112x1x1]
%onnx::Conv_852[FLOAT, 336]
%onnx::Conv_854[FLOAT, 336x1x3x3]
%onnx::Conv_857[FLOAT, 112x336x1x1]
%onnx::Conv_860[FLOAT, 112x112x1x1]
%onnx::Conv_863[FLOAT, 112x1x5x5]
%onnx::Conv_866[FLOAT, 112x112x1x1]
%onnx::Conv_869[FLOAT, 112x56x1x1]
%onnx::Conv_872[FLOAT, 112x1x3x3]
%onnx::Conv_875[FLOAT, 184x56x1x1]
%onnx::Conv_876[FLOAT, 184]
%onnx::Conv_878[FLOAT, 552x184x1x1]
%onnx::Conv_879[FLOAT, 552]
%onnx::Conv_881[FLOAT, 552x1x3x3]
%onnx::Conv_884[FLOAT, 184x552x1x1]
%onnx::Conv_887[FLOAT, 1104x184x1x1]
%onnx::Conv_888[FLOAT, 1104]
%onnx::Conv_890[FLOAT, 1104x1x5x5]
%onnx::Conv_893[FLOAT, 184x1104x1x1]
%onnx::Conv_896[FLOAT, 184x92x1x1]
%onnx::Conv_899[FLOAT, 184x1x5x5]
%onnx::Conv_902[FLOAT, 184x92x1x1]
%onnx::Conv_905[FLOAT, 552x184x1x1]
%onnx::Conv_908[FLOAT, 552x1x3x3]
%onnx::Conv_911[FLOAT, 352x552x1x1]
%onnx::Conv_912[FLOAT, 352]
%onnx::Conv_914[FLOAT, 1504x352x1x1]
%onnx::Conv_915[FLOAT, 1504]
) {
%onnx::Conv_909 = Identity(%onnx::Conv_879)
%onnx::Conv_906 = Identity(%onnx::Conv_879)
%onnx::Conv_903 = Identity(%onnx::Conv_876)
%onnx::Conv_900 = Identity(%onnx::Conv_876)
%onnx::Conv_897 = Identity(%onnx::Conv_876)
%onnx::Conv_894 = Identity(%onnx::Conv_876)
%onnx::Conv_891 = Identity(%onnx::Conv_888)
%onnx::Conv_885 = Identity(%onnx::Conv_876)
%onnx::Conv_882 = Identity(%onnx::Conv_879)
%onnx::Conv_873 = Identity(%onnx::Conv_840)
%onnx::Conv_870 = Identity(%onnx::Conv_840)
%onnx::Conv_867 = Identity(%onnx::Conv_840)
%onnx::Conv_864 = Identity(%onnx::Conv_840)
%onnx::Conv_861 = Identity(%onnx::Conv_840)
%onnx::Conv_858 = Identity(%onnx::Conv_840)
%onnx::Conv_855 = Identity(%onnx::Conv_852)
%onnx::Conv_849 = Identity(%onnx::Conv_840)
%onnx::Conv_846 = Identity(%onnx::Conv_843)
%onnx::Conv_837 = Identity(%onnx::Conv_807)
%onnx::Conv_834 = Identity(%onnx::Conv_807)
%onnx::Conv_831 = Identity(%onnx::Conv_804)
%onnx::Conv_828 = Identity(%onnx::Conv_825)
%onnx::Conv_822 = Identity(%onnx::Conv_804)
%onnx::Conv_819 = Identity(%onnx::Conv_804)
%onnx::Conv_816 = Identity(%onnx::Conv_804)
%onnx::Conv_813 = Identity(%onnx::Conv_804)
%onnx::Conv_810 = Identity(%onnx::Conv_807)
%onnx::Conv_801 = Identity(%onnx::Conv_768)
%onnx::Conv_798 = Identity(%onnx::Conv_768)
%onnx::Conv_795 = Identity(%onnx::Conv_768)
%onnx::Conv_792 = Identity(%onnx::Conv_768)
%onnx::Conv_789 = Identity(%onnx::Conv_768)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_768)
%onnx::Conv_780 = Identity(%onnx::Conv_768)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_768)
%onnx::Conv_771 = Identity(%onnx::Conv_768)
%onnx::Conv_765 = Identity(%onnx::Conv_744)
%onnx::Conv_762 = Identity(%onnx::Conv_744)
%onnx::Conv_759 = Identity(%onnx::Conv_732)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_750 = Identity(%onnx::Conv_732)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_732)
%onnx::Conv_735 = Identity(%onnx::Conv_732)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_714)
%onnx::Conv_717 = Identity(%onnx::Conv_714)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_713, %onnx::Conv_714)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_872, %onnx::Conv_873)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_875, %onnx::Conv_876)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_878, %onnx::Conv_879)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_881, %onnx::Conv_882)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_884, %onnx::Conv_885)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_887, %onnx::Conv_888)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_890, %onnx::Conv_891)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_893, %onnx::Conv_894)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_896, %onnx::Conv_897)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_899, %onnx::Conv_900)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_902, %onnx::Conv_903)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_905, %onnx::Conv_906)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_908, %onnx::Conv_909)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_911, %onnx::Conv_912)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_914, %onnx::Conv_915)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%711 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %711
}
|
val_accuracy
| 0 | 82,487,424 | 2,178,036 |
{'zcp_synflow': 84.43782217578548, 'zcp_zen': 74.56391906738281, 'zcp_epe_nas': 28.97624773918945, 'zcp_fisher': 0.15244294703006744, 'zcp_flops': 82487424.0, 'zcp_grad_norm': 26.920394897460938, 'zcp_grasp': -0.11804008483886719, 'zcp_jacov': -16.060512617248314, 'zcp_l2_norm': 682.94482421875, 'zcp_nwot': 214.1157188197237, 'zcp_params': 2178036.0, 'zcp_plain': 0.001358098816126585, 'zcp_snip': 49.80146026611328, 'lat_1080ti_1': 0.7775898490040473, 'lat_1080ti_32': 0.7675764622700787, 'lat_1080ti_64': 0.6497638343016943, 'lat_2080ti_1': 0.8320329775275328, 'lat_2080ti_32': 0.7728421186692761, 'lat_2080ti_64': 0.6509283469894512, 'lat_essential_ph_1': 0.5471698113207547, 'lat_eyeriss': 0.5912016577118441, 'lat_fpga': 0.6439109826932782, 'lat_gold_6226': 0.4748741004779205, 'lat_gold_6240': 0.6403675134159186, 'lat_pixel2': 0.43478260869565216, 'lat_pixel3': 0.6256357038951421, 'lat_raspi4': 0.673065136069794, 'lat_samsung_a50': 0.2736842105263158, 'lat_samsung_s7': 0.2283464566929134, 'lat_silver_4114': 0.6602674317786876, 'lat_silver_4210r': 0.7102774983706711, 'lat_titan_rtx_1': 0.7826274233940573, 'lat_titan_rtx_32': 0.7621304415479992, 'lat_titan_rtx_64': 0.683411391666467, 'lat_titanx_1': 0.4222110920624317, 'lat_titanx_32': 0.7277711803065343, 'lat_titanx_64': 0.6713335550960117, 'lat_titanxp_1': 0.7651711634033709, 'lat_titanxp_32': 0.7585454908794671, 'lat_titanxp_64': 0.6817924279168471}
| |
FBNet_2912
|
FBNet
|
2912
|
2912
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_713[FLOAT, 16x3x3x3]
%onnx::Conv_714[FLOAT, 16]
%onnx::Conv_716[FLOAT, 96x16x1x1]
%onnx::Conv_717[FLOAT, 96]
%onnx::Conv_719[FLOAT, 96x1x5x5]
%onnx::Conv_722[FLOAT, 16x96x1x1]
%onnx::Conv_725[FLOAT, 16x16x1x1]
%onnx::Conv_728[FLOAT, 16x1x3x3]
%onnx::Conv_731[FLOAT, 24x16x1x1]
%onnx::Conv_732[FLOAT, 24]
%onnx::Conv_734[FLOAT, 72x24x1x1]
%onnx::Conv_735[FLOAT, 72]
%onnx::Conv_737[FLOAT, 72x1x5x5]
%onnx::Conv_740[FLOAT, 24x72x1x1]
%onnx::Conv_743[FLOAT, 72x24x1x1]
%onnx::Conv_746[FLOAT, 72x1x3x3]
%onnx::Conv_749[FLOAT, 24x72x1x1]
%onnx::Conv_752[FLOAT, 24x24x1x1]
%onnx::Conv_755[FLOAT, 24x1x3x3]
%onnx::Conv_758[FLOAT, 24x24x1x1]
%onnx::Conv_761[FLOAT, 72x24x1x1]
%onnx::Conv_764[FLOAT, 72x1x3x3]
%onnx::Conv_767[FLOAT, 32x72x1x1]
%onnx::Conv_768[FLOAT, 32]
%onnx::Conv_770[FLOAT, 96x32x1x1]
%onnx::Conv_773[FLOAT, 96x1x5x5]
%onnx::Conv_776[FLOAT, 32x96x1x1]
%onnx::Conv_779[FLOAT, 32x16x1x1]
%onnx::Conv_782[FLOAT, 32x1x5x5]
%onnx::Conv_785[FLOAT, 32x16x1x1]
%onnx::Conv_788[FLOAT, 32x16x1x1]
%onnx::Conv_791[FLOAT, 32x1x5x5]
%onnx::Conv_794[FLOAT, 64x16x1x1]
%onnx::Conv_795[FLOAT, 64]
%onnx::Conv_797[FLOAT, 64x64x1x1]
%onnx::Conv_800[FLOAT, 64x1x5x5]
%onnx::Conv_803[FLOAT, 64x64x1x1]
%onnx::Conv_806[FLOAT, 384x64x1x1]
%onnx::Conv_807[FLOAT, 384]
%onnx::Conv_809[FLOAT, 384x1x5x5]
%onnx::Conv_812[FLOAT, 64x384x1x1]
%onnx::Conv_815[FLOAT, 64x32x1x1]
%onnx::Conv_818[FLOAT, 64x1x5x5]
%onnx::Conv_821[FLOAT, 112x32x1x1]
%onnx::Conv_822[FLOAT, 112]
%onnx::Conv_824[FLOAT, 112x56x1x1]
%onnx::Conv_827[FLOAT, 112x1x3x3]
%onnx::Conv_830[FLOAT, 112x56x1x1]
%onnx::Conv_833[FLOAT, 336x112x1x1]
%onnx::Conv_834[FLOAT, 336]
%onnx::Conv_836[FLOAT, 336x1x3x3]
%onnx::Conv_839[FLOAT, 112x336x1x1]
%onnx::Conv_842[FLOAT, 672x112x1x1]
%onnx::Conv_843[FLOAT, 672]
%onnx::Conv_845[FLOAT, 672x1x5x5]
%onnx::Conv_848[FLOAT, 112x672x1x1]
%onnx::Conv_851[FLOAT, 112x112x1x1]
%onnx::Conv_854[FLOAT, 112x1x3x3]
%onnx::Conv_857[FLOAT, 184x112x1x1]
%onnx::Conv_858[FLOAT, 184]
%onnx::Conv_860[FLOAT, 184x92x1x1]
%onnx::Conv_863[FLOAT, 184x1x5x5]
%onnx::Conv_866[FLOAT, 184x92x1x1]
%onnx::Conv_869[FLOAT, 184x92x1x1]
%onnx::Conv_872[FLOAT, 184x1x5x5]
%onnx::Conv_875[FLOAT, 184x92x1x1]
%onnx::Conv_878[FLOAT, 184x92x1x1]
%onnx::Conv_881[FLOAT, 184x1x3x3]
%onnx::Conv_884[FLOAT, 184x92x1x1]
%onnx::Conv_887[FLOAT, 184x92x1x1]
%onnx::Conv_890[FLOAT, 184x1x3x3]
%onnx::Conv_893[FLOAT, 352x92x1x1]
%onnx::Conv_894[FLOAT, 352]
%onnx::Conv_896[FLOAT, 1504x352x1x1]
%onnx::Conv_897[FLOAT, 1504]
) {
%onnx::Conv_891 = Identity(%onnx::Conv_858)
%onnx::Conv_888 = Identity(%onnx::Conv_858)
%onnx::Conv_885 = Identity(%onnx::Conv_858)
%onnx::Conv_882 = Identity(%onnx::Conv_858)
%onnx::Conv_879 = Identity(%onnx::Conv_858)
%onnx::Conv_876 = Identity(%onnx::Conv_858)
%onnx::Conv_873 = Identity(%onnx::Conv_858)
%onnx::Conv_870 = Identity(%onnx::Conv_858)
%onnx::Conv_867 = Identity(%onnx::Conv_858)
%onnx::Conv_864 = Identity(%onnx::Conv_858)
%onnx::Conv_861 = Identity(%onnx::Conv_858)
%onnx::Conv_855 = Identity(%onnx::Conv_822)
%onnx::Conv_852 = Identity(%onnx::Conv_822)
%onnx::Conv_849 = Identity(%onnx::Conv_822)
%onnx::Conv_846 = Identity(%onnx::Conv_843)
%onnx::Conv_840 = Identity(%onnx::Conv_822)
%onnx::Conv_837 = Identity(%onnx::Conv_834)
%onnx::Conv_831 = Identity(%onnx::Conv_822)
%onnx::Conv_828 = Identity(%onnx::Conv_822)
%onnx::Conv_825 = Identity(%onnx::Conv_822)
%onnx::Conv_819 = Identity(%onnx::Conv_795)
%onnx::Conv_816 = Identity(%onnx::Conv_795)
%onnx::Conv_813 = Identity(%onnx::Conv_795)
%onnx::Conv_810 = Identity(%onnx::Conv_807)
%onnx::Conv_804 = Identity(%onnx::Conv_795)
%onnx::Conv_801 = Identity(%onnx::Conv_795)
%onnx::Conv_798 = Identity(%onnx::Conv_795)
%onnx::Conv_792 = Identity(%onnx::Conv_768)
%onnx::Conv_789 = Identity(%onnx::Conv_768)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_768)
%onnx::Conv_780 = Identity(%onnx::Conv_768)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_717)
%onnx::Conv_771 = Identity(%onnx::Conv_717)
%onnx::Conv_765 = Identity(%onnx::Conv_735)
%onnx::Conv_762 = Identity(%onnx::Conv_735)
%onnx::Conv_759 = Identity(%onnx::Conv_732)
%onnx::Conv_756 = Identity(%onnx::Conv_732)
%onnx::Conv_753 = Identity(%onnx::Conv_732)
%onnx::Conv_750 = Identity(%onnx::Conv_732)
%onnx::Conv_747 = Identity(%onnx::Conv_735)
%onnx::Conv_744 = Identity(%onnx::Conv_735)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_729 = Identity(%onnx::Conv_714)
%onnx::Conv_726 = Identity(%onnx::Conv_714)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_713, %onnx::Conv_714)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_872, %onnx::Conv_873)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_875, %onnx::Conv_876)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_878, %onnx::Conv_879)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_881, %onnx::Conv_882)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_884, %onnx::Conv_885)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_887, %onnx::Conv_888)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_890, %onnx::Conv_891)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_893, %onnx::Conv_894)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_896, %onnx::Conv_897)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%711 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %711
}
|
val_accuracy
| 0 | 60,045,696 | 1,266,084 |
{'zcp_synflow': 71.94476337701404, 'zcp_zen': 63.70613098144531, 'zcp_epe_nas': 35.3216359760123, 'zcp_fisher': 0.13567420840263367, 'zcp_flops': 60045696.0, 'zcp_grad_norm': 25.941770553588867, 'zcp_grasp': -0.6238784790039062, 'zcp_jacov': -16.05877924273549, 'zcp_l2_norm': 532.0971069335938, 'zcp_nwot': 211.91344691747855, 'zcp_params': 1266084.0, 'zcp_plain': 0.0019201562972739339, 'zcp_snip': 40.48036193847656, 'lat_1080ti_1': 0.6831182505780456, 'lat_1080ti_32': 0.56496235447306, 'lat_1080ti_64': 0.4965771637160825, 'lat_2080ti_1': 0.6845850924497547, 'lat_2080ti_32': 0.6053013056148236, 'lat_2080ti_64': 0.5185055571447282, 'lat_essential_ph_1': 0.3018867924528302, 'lat_eyeriss': 0.3263986943685475, 'lat_fpga': 0.34102850809922913, 'lat_gold_6226': 0.18919731495149106, 'lat_gold_6240': 0.3438872657974809, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.3838952212644901, 'lat_raspi4': 0.35240691304215593, 'lat_samsung_a50': 0.1368421052631579, 'lat_samsung_s7': 0.12598425196850394, 'lat_silver_4114': 0.48005442465279974, 'lat_silver_4210r': 0.3955733196792823, 'lat_titan_rtx_1': 0.641141833992723, 'lat_titan_rtx_32': 0.5883975476884544, 'lat_titan_rtx_64': 0.5411202368515784, 'lat_titanx_1': 0.3346442065914383, 'lat_titanx_32': 0.5663851091258528, 'lat_titanx_64': 0.5441017359089758, 'lat_titanxp_1': 0.5859177175531924, 'lat_titanxp_32': 0.5960962937204705, 'lat_titanxp_64': 0.5225451204808248}
| |
FBNet_4226
|
FBNet
|
4226
|
4226
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_587[FLOAT, 16x3x3x3]
%onnx::Conv_588[FLOAT, 16]
%onnx::Conv_590[FLOAT, 16x16x1x1]
%onnx::Conv_593[FLOAT, 16x1x5x5]
%onnx::Conv_596[FLOAT, 16x16x1x1]
%onnx::Conv_599[FLOAT, 24x16x1x1]
%onnx::Conv_600[FLOAT, 24]
%onnx::Conv_602[FLOAT, 24x24x1x1]
%onnx::Conv_605[FLOAT, 24x1x3x3]
%onnx::Conv_608[FLOAT, 24x24x1x1]
%onnx::Conv_611[FLOAT, 144x24x1x1]
%onnx::Conv_612[FLOAT, 144]
%onnx::Conv_614[FLOAT, 144x1x5x5]
%onnx::Conv_617[FLOAT, 24x144x1x1]
%onnx::Conv_620[FLOAT, 72x24x1x1]
%onnx::Conv_621[FLOAT, 72]
%onnx::Conv_623[FLOAT, 72x1x5x5]
%onnx::Conv_626[FLOAT, 24x72x1x1]
%onnx::Conv_629[FLOAT, 144x24x1x1]
%onnx::Conv_632[FLOAT, 144x1x5x5]
%onnx::Conv_635[FLOAT, 32x144x1x1]
%onnx::Conv_636[FLOAT, 32]
%onnx::Conv_638[FLOAT, 96x32x1x1]
%onnx::Conv_639[FLOAT, 96]
%onnx::Conv_641[FLOAT, 96x1x3x3]
%onnx::Conv_644[FLOAT, 32x96x1x1]
%onnx::Conv_647[FLOAT, 192x32x1x1]
%onnx::Conv_648[FLOAT, 192]
%onnx::Conv_650[FLOAT, 192x1x5x5]
%onnx::Conv_653[FLOAT, 32x192x1x1]
%onnx::Conv_656[FLOAT, 96x32x1x1]
%onnx::Conv_659[FLOAT, 96x1x3x3]
%onnx::Conv_662[FLOAT, 32x96x1x1]
%onnx::Conv_665[FLOAT, 32x32x1x1]
%onnx::Conv_668[FLOAT, 32x1x3x3]
%onnx::Conv_671[FLOAT, 64x32x1x1]
%onnx::Conv_672[FLOAT, 64]
%onnx::Conv_674[FLOAT, 384x64x1x1]
%onnx::Conv_675[FLOAT, 384]
%onnx::Conv_677[FLOAT, 384x1x3x3]
%onnx::Conv_680[FLOAT, 64x384x1x1]
%onnx::Conv_683[FLOAT, 384x64x1x1]
%onnx::Conv_686[FLOAT, 384x1x5x5]
%onnx::Conv_689[FLOAT, 64x384x1x1]
%onnx::Conv_692[FLOAT, 192x64x1x1]
%onnx::Conv_695[FLOAT, 192x1x3x3]
%onnx::Conv_698[FLOAT, 112x192x1x1]
%onnx::Conv_699[FLOAT, 112]
%onnx::Conv_701[FLOAT, 336x112x1x1]
%onnx::Conv_702[FLOAT, 336]
%onnx::Conv_704[FLOAT, 336x1x3x3]
%onnx::Conv_707[FLOAT, 112x336x1x1]
%onnx::Conv_710[FLOAT, 112x56x1x1]
%onnx::Conv_713[FLOAT, 112x1x3x3]
%onnx::Conv_716[FLOAT, 112x56x1x1]
%onnx::Conv_719[FLOAT, 112x112x1x1]
%onnx::Conv_722[FLOAT, 112x1x5x5]
%onnx::Conv_725[FLOAT, 184x112x1x1]
%onnx::Conv_726[FLOAT, 184]
%onnx::Conv_728[FLOAT, 184x92x1x1]
%onnx::Conv_731[FLOAT, 184x1x3x3]
%onnx::Conv_734[FLOAT, 184x92x1x1]
%onnx::Conv_737[FLOAT, 184x92x1x1]
%onnx::Conv_740[FLOAT, 184x1x3x3]
%onnx::Conv_743[FLOAT, 184x92x1x1]
%onnx::Conv_746[FLOAT, 184x184x1x1]
%onnx::Conv_749[FLOAT, 184x1x3x3]
%onnx::Conv_752[FLOAT, 184x184x1x1]
%onnx::Conv_755[FLOAT, 352x184x1x1]
%onnx::Conv_756[FLOAT, 352]
%onnx::Conv_758[FLOAT, 1504x352x1x1]
%onnx::Conv_759[FLOAT, 1504]
) {
%onnx::Conv_753 = Identity(%onnx::Conv_726)
%onnx::Conv_750 = Identity(%onnx::Conv_726)
%onnx::Conv_747 = Identity(%onnx::Conv_726)
%onnx::Conv_744 = Identity(%onnx::Conv_726)
%onnx::Conv_741 = Identity(%onnx::Conv_726)
%onnx::Conv_738 = Identity(%onnx::Conv_726)
%onnx::Conv_735 = Identity(%onnx::Conv_726)
%onnx::Conv_732 = Identity(%onnx::Conv_726)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_723 = Identity(%onnx::Conv_699)
%onnx::Conv_720 = Identity(%onnx::Conv_699)
%onnx::Conv_717 = Identity(%onnx::Conv_699)
%onnx::Conv_714 = Identity(%onnx::Conv_699)
%onnx::Conv_711 = Identity(%onnx::Conv_699)
%onnx::Conv_708 = Identity(%onnx::Conv_699)
%onnx::Conv_705 = Identity(%onnx::Conv_702)
%onnx::Conv_696 = Identity(%onnx::Conv_648)
%onnx::Conv_693 = Identity(%onnx::Conv_648)
%onnx::Conv_690 = Identity(%onnx::Conv_672)
%onnx::Conv_687 = Identity(%onnx::Conv_675)
%onnx::Conv_684 = Identity(%onnx::Conv_675)
%onnx::Conv_681 = Identity(%onnx::Conv_672)
%onnx::Conv_678 = Identity(%onnx::Conv_675)
%onnx::Conv_669 = Identity(%onnx::Conv_636)
%onnx::Conv_666 = Identity(%onnx::Conv_636)
%onnx::Conv_663 = Identity(%onnx::Conv_636)
%onnx::Conv_660 = Identity(%onnx::Conv_639)
%onnx::Conv_657 = Identity(%onnx::Conv_639)
%onnx::Conv_654 = Identity(%onnx::Conv_636)
%onnx::Conv_651 = Identity(%onnx::Conv_648)
%onnx::Conv_645 = Identity(%onnx::Conv_636)
%onnx::Conv_642 = Identity(%onnx::Conv_639)
%onnx::Conv_633 = Identity(%onnx::Conv_612)
%onnx::Conv_630 = Identity(%onnx::Conv_612)
%onnx::Conv_627 = Identity(%onnx::Conv_600)
%onnx::Conv_624 = Identity(%onnx::Conv_621)
%onnx::Conv_618 = Identity(%onnx::Conv_600)
%onnx::Conv_615 = Identity(%onnx::Conv_612)
%onnx::Conv_609 = Identity(%onnx::Conv_600)
%onnx::Conv_606 = Identity(%onnx::Conv_600)
%onnx::Conv_603 = Identity(%onnx::Conv_600)
%onnx::Conv_597 = Identity(%onnx::Conv_588)
%onnx::Conv_594 = Identity(%onnx::Conv_588)
%onnx::Conv_591 = Identity(%onnx::Conv_588)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_587, %onnx::Conv_588)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_590, %onnx::Conv_591)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_593, %onnx::Conv_594)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_596, %onnx::Conv_597)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_599, %onnx::Conv_600)
%/cells.1/relu/Relu_output_0 = Relu(%/cells.1/conv/Conv_output_0)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/relu/Relu_output_0, %onnx::Conv_602, %onnx::Conv_603)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_605, %onnx::Conv_606)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/relu/Relu_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%585 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %585
}
|
val_accuracy
| 0 | 65,007,744 | 1,243,036 |
{'zcp_synflow': 74.8614613098336, 'zcp_zen': 62.68047332763672, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.1129792258143425, 'zcp_flops': 65007744.0, 'zcp_grad_norm': 19.930503845214844, 'zcp_grasp': 0.03329277038574219, 'zcp_jacov': -16.060899837130318, 'zcp_l2_norm': 552.54052734375, 'zcp_nwot': 214.65757833932602, 'zcp_params': 1243036.0, 'zcp_plain': -0.0035233781673014164, 'zcp_snip': 33.83395767211914, 'lat_1080ti_1': 0.46606354792579613, 'lat_1080ti_32': 0.5031823677899941, 'lat_1080ti_64': 0.4958108172632888, 'lat_2080ti_1': 0.4652936836187114, 'lat_2080ti_32': 0.4835307617490174, 'lat_2080ti_64': 0.50466459376703, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.42125868740028993, 'lat_fpga': 0.33653608288227665, 'lat_gold_6226': 0.2186243249441841, 'lat_gold_6240': 0.3311635266338261, 'lat_pixel2': 0.30434782608695654, 'lat_pixel3': 0.44710641605864154, 'lat_raspi4': 0.41481655947119345, 'lat_samsung_a50': 0.15789473684210525, 'lat_samsung_s7': 0.07086614173228346, 'lat_silver_4114': 0.2733464708731741, 'lat_silver_4210r': 0.2956504035854428, 'lat_titan_rtx_1': 0.4397064297504909, 'lat_titan_rtx_32': 0.44083277820984385, 'lat_titan_rtx_64': 0.5044387082820053, 'lat_titanx_1': 0.22203576709385803, 'lat_titanx_32': 0.49155728381328634, 'lat_titanx_64': 0.49034141048639346, 'lat_titanxp_1': 0.43127488137158254, 'lat_titanxp_32': 0.5022091488254922, 'lat_titanxp_64': 0.5102889720382575}
| |
FBNet_2049
|
FBNet
|
2049
|
2049
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_644[FLOAT, 16x3x3x3]
%onnx::Conv_645[FLOAT, 16]
%onnx::Conv_647[FLOAT, 16x8x1x1]
%onnx::Conv_650[FLOAT, 16x1x3x3]
%onnx::Conv_653[FLOAT, 16x8x1x1]
%onnx::Conv_656[FLOAT, 16x16x1x1]
%onnx::Conv_659[FLOAT, 16x1x5x5]
%onnx::Conv_662[FLOAT, 24x16x1x1]
%onnx::Conv_663[FLOAT, 24]
%onnx::Conv_665[FLOAT, 72x24x1x1]
%onnx::Conv_666[FLOAT, 72]
%onnx::Conv_668[FLOAT, 72x1x3x3]
%onnx::Conv_671[FLOAT, 24x72x1x1]
%onnx::Conv_674[FLOAT, 144x24x1x1]
%onnx::Conv_675[FLOAT, 144]
%onnx::Conv_677[FLOAT, 144x1x3x3]
%onnx::Conv_680[FLOAT, 24x144x1x1]
%onnx::Conv_683[FLOAT, 24x12x1x1]
%onnx::Conv_686[FLOAT, 24x1x5x5]
%onnx::Conv_689[FLOAT, 24x12x1x1]
%onnx::Conv_692[FLOAT, 24x24x1x1]
%onnx::Conv_695[FLOAT, 24x1x5x5]
%onnx::Conv_698[FLOAT, 32x24x1x1]
%onnx::Conv_699[FLOAT, 32]
%onnx::Conv_701[FLOAT, 32x32x1x1]
%onnx::Conv_704[FLOAT, 32x1x3x3]
%onnx::Conv_707[FLOAT, 32x32x1x1]
%onnx::Conv_710[FLOAT, 192x32x1x1]
%onnx::Conv_711[FLOAT, 192]
%onnx::Conv_713[FLOAT, 192x1x3x3]
%onnx::Conv_716[FLOAT, 32x192x1x1]
%onnx::Conv_719[FLOAT, 32x32x1x1]
%onnx::Conv_722[FLOAT, 32x1x3x3]
%onnx::Conv_725[FLOAT, 32x32x1x1]
%onnx::Conv_728[FLOAT, 64x32x1x1]
%onnx::Conv_729[FLOAT, 64]
%onnx::Conv_731[FLOAT, 192x64x1x1]
%onnx::Conv_734[FLOAT, 192x1x3x3]
%onnx::Conv_737[FLOAT, 64x192x1x1]
%onnx::Conv_740[FLOAT, 64x32x1x1]
%onnx::Conv_743[FLOAT, 64x1x5x5]
%onnx::Conv_746[FLOAT, 64x32x1x1]
%onnx::Conv_749[FLOAT, 64x32x1x1]
%onnx::Conv_752[FLOAT, 64x1x5x5]
%onnx::Conv_755[FLOAT, 64x32x1x1]
%onnx::Conv_758[FLOAT, 112x64x1x1]
%onnx::Conv_759[FLOAT, 112]
%onnx::Conv_761[FLOAT, 112x56x1x1]
%onnx::Conv_764[FLOAT, 112x1x3x3]
%onnx::Conv_767[FLOAT, 112x56x1x1]
%onnx::Conv_770[FLOAT, 112x112x1x1]
%onnx::Conv_773[FLOAT, 112x1x5x5]
%onnx::Conv_776[FLOAT, 112x112x1x1]
%onnx::Conv_779[FLOAT, 336x112x1x1]
%onnx::Conv_780[FLOAT, 336]
%onnx::Conv_782[FLOAT, 336x1x3x3]
%onnx::Conv_785[FLOAT, 184x336x1x1]
%onnx::Conv_786[FLOAT, 184]
%onnx::Conv_788[FLOAT, 184x92x1x1]
%onnx::Conv_791[FLOAT, 184x1x5x5]
%onnx::Conv_794[FLOAT, 184x92x1x1]
%onnx::Conv_797[FLOAT, 184x184x1x1]
%onnx::Conv_800[FLOAT, 184x1x3x3]
%onnx::Conv_803[FLOAT, 184x184x1x1]
%onnx::Conv_806[FLOAT, 184x184x1x1]
%onnx::Conv_809[FLOAT, 184x1x3x3]
%onnx::Conv_812[FLOAT, 352x184x1x1]
%onnx::Conv_813[FLOAT, 352]
%onnx::Conv_815[FLOAT, 1504x352x1x1]
%onnx::Conv_816[FLOAT, 1504]
) {
%onnx::Conv_810 = Identity(%onnx::Conv_786)
%onnx::Conv_807 = Identity(%onnx::Conv_786)
%onnx::Conv_804 = Identity(%onnx::Conv_786)
%onnx::Conv_801 = Identity(%onnx::Conv_786)
%onnx::Conv_798 = Identity(%onnx::Conv_786)
%onnx::Conv_795 = Identity(%onnx::Conv_786)
%onnx::Conv_792 = Identity(%onnx::Conv_786)
%onnx::Conv_789 = Identity(%onnx::Conv_786)
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_777 = Identity(%onnx::Conv_759)
%onnx::Conv_774 = Identity(%onnx::Conv_759)
%onnx::Conv_771 = Identity(%onnx::Conv_759)
%onnx::Conv_768 = Identity(%onnx::Conv_759)
%onnx::Conv_765 = Identity(%onnx::Conv_759)
%onnx::Conv_762 = Identity(%onnx::Conv_759)
%onnx::Conv_756 = Identity(%onnx::Conv_729)
%onnx::Conv_753 = Identity(%onnx::Conv_729)
%onnx::Conv_750 = Identity(%onnx::Conv_729)
%onnx::Conv_747 = Identity(%onnx::Conv_729)
%onnx::Conv_744 = Identity(%onnx::Conv_729)
%onnx::Conv_741 = Identity(%onnx::Conv_729)
%onnx::Conv_738 = Identity(%onnx::Conv_729)
%onnx::Conv_735 = Identity(%onnx::Conv_711)
%onnx::Conv_732 = Identity(%onnx::Conv_711)
%onnx::Conv_726 = Identity(%onnx::Conv_699)
%onnx::Conv_723 = Identity(%onnx::Conv_699)
%onnx::Conv_720 = Identity(%onnx::Conv_699)
%onnx::Conv_717 = Identity(%onnx::Conv_699)
%onnx::Conv_714 = Identity(%onnx::Conv_711)
%onnx::Conv_708 = Identity(%onnx::Conv_699)
%onnx::Conv_705 = Identity(%onnx::Conv_699)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%onnx::Conv_696 = Identity(%onnx::Conv_663)
%onnx::Conv_693 = Identity(%onnx::Conv_663)
%onnx::Conv_690 = Identity(%onnx::Conv_663)
%onnx::Conv_687 = Identity(%onnx::Conv_663)
%onnx::Conv_684 = Identity(%onnx::Conv_663)
%onnx::Conv_681 = Identity(%onnx::Conv_663)
%onnx::Conv_678 = Identity(%onnx::Conv_675)
%onnx::Conv_672 = Identity(%onnx::Conv_663)
%onnx::Conv_669 = Identity(%onnx::Conv_666)
%onnx::Conv_660 = Identity(%onnx::Conv_645)
%onnx::Conv_657 = Identity(%onnx::Conv_645)
%onnx::Conv_654 = Identity(%onnx::Conv_645)
%onnx::Conv_651 = Identity(%onnx::Conv_645)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_644, %onnx::Conv_645)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%642 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %642
}
|
val_accuracy
| 0 | 45,125,504 | 1,129,284 |
{'zcp_synflow': 69.90182079921912, 'zcp_zen': 57.12721252441406, 'zcp_epe_nas': 22.553798279535112, 'zcp_fisher': 0.05553493648767471, 'zcp_flops': 45125504.0, 'zcp_grad_norm': 18.87816619873047, 'zcp_grasp': -0.023353099822998047, 'zcp_jacov': -16.045885316774648, 'zcp_l2_norm': 478.0677795410156, 'zcp_nwot': 209.63876737083976, 'zcp_params': 1129284.0, 'zcp_plain': 0.004259450826793909, 'zcp_snip': 27.022727966308594, 'lat_1080ti_1': 0.4432039263039743, 'lat_1080ti_32': 0.5232916420257139, 'lat_1080ti_64': 0.3561304234652709, 'lat_2080ti_1': 0.6192373743000812, 'lat_2080ti_32': 0.48738764085539327, 'lat_2080ti_64': 0.39116603901341074, 'lat_essential_ph_1': 0.1320754716981132, 'lat_eyeriss': 0.17699374690554343, 'lat_fpga': 0.16194500242432638, 'lat_gold_6226': 0.05266320625702926, 'lat_gold_6240': 0.2188767292308029, 'lat_pixel2': 0.08695652173913043, 'lat_pixel3': 0.1792232342781145, 'lat_raspi4': 0.19433950884759182, 'lat_samsung_a50': 0.06315789473684211, 'lat_samsung_s7': 0.05511811023622047, 'lat_silver_4114': 0.2649728202726456, 'lat_silver_4210r': 0.26964031633290614, 'lat_titan_rtx_1': 0.4948417309693694, 'lat_titan_rtx_32': 0.45184640737668286, 'lat_titan_rtx_64': 0.40676511362842416, 'lat_titanx_1': 0.2483681781211684, 'lat_titanx_32': 0.419764866823364, 'lat_titanx_64': 0.34850132760640756, 'lat_titanxp_1': 0.4679690415165846, 'lat_titanxp_32': 0.4330876559445033, 'lat_titanxp_64': 0.36988660207300994}
| |
FBNet_2554
|
FBNet
|
2554
|
2554
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_759[FLOAT, 16x3x3x3]
%onnx::Conv_760[FLOAT, 16]
%onnx::Conv_762[FLOAT, 96x16x1x1]
%onnx::Conv_763[FLOAT, 96]
%onnx::Conv_765[FLOAT, 96x1x3x3]
%onnx::Conv_768[FLOAT, 16x96x1x1]
%onnx::Conv_771[FLOAT, 96x16x1x1]
%onnx::Conv_774[FLOAT, 96x1x5x5]
%onnx::Conv_777[FLOAT, 24x96x1x1]
%onnx::Conv_778[FLOAT, 24]
%onnx::Conv_780[FLOAT, 24x12x1x1]
%onnx::Conv_783[FLOAT, 24x1x5x5]
%onnx::Conv_786[FLOAT, 24x12x1x1]
%onnx::Conv_789[FLOAT, 24x12x1x1]
%onnx::Conv_792[FLOAT, 24x1x5x5]
%onnx::Conv_795[FLOAT, 24x12x1x1]
%onnx::Conv_798[FLOAT, 24x12x1x1]
%onnx::Conv_801[FLOAT, 24x1x3x3]
%onnx::Conv_804[FLOAT, 32x12x1x1]
%onnx::Conv_805[FLOAT, 32]
%onnx::Conv_807[FLOAT, 32x16x1x1]
%onnx::Conv_810[FLOAT, 32x1x3x3]
%onnx::Conv_813[FLOAT, 32x16x1x1]
%onnx::Conv_816[FLOAT, 32x16x1x1]
%onnx::Conv_819[FLOAT, 32x1x5x5]
%onnx::Conv_822[FLOAT, 32x16x1x1]
%onnx::Conv_825[FLOAT, 32x16x1x1]
%onnx::Conv_828[FLOAT, 32x1x3x3]
%onnx::Conv_831[FLOAT, 32x16x1x1]
%onnx::Conv_834[FLOAT, 192x32x1x1]
%onnx::Conv_835[FLOAT, 192]
%onnx::Conv_837[FLOAT, 192x1x3x3]
%onnx::Conv_840[FLOAT, 64x192x1x1]
%onnx::Conv_841[FLOAT, 64]
%onnx::Conv_843[FLOAT, 64x32x1x1]
%onnx::Conv_846[FLOAT, 64x1x3x3]
%onnx::Conv_849[FLOAT, 64x32x1x1]
%onnx::Conv_852[FLOAT, 64x64x1x1]
%onnx::Conv_855[FLOAT, 64x1x5x5]
%onnx::Conv_858[FLOAT, 64x64x1x1]
%onnx::Conv_861[FLOAT, 192x64x1x1]
%onnx::Conv_864[FLOAT, 192x1x5x5]
%onnx::Conv_867[FLOAT, 64x192x1x1]
%onnx::Conv_870[FLOAT, 64x64x1x1]
%onnx::Conv_873[FLOAT, 64x1x3x3]
%onnx::Conv_876[FLOAT, 112x64x1x1]
%onnx::Conv_877[FLOAT, 112]
%onnx::Conv_879[FLOAT, 336x112x1x1]
%onnx::Conv_880[FLOAT, 336]
%onnx::Conv_882[FLOAT, 336x1x3x3]
%onnx::Conv_885[FLOAT, 112x336x1x1]
%onnx::Conv_888[FLOAT, 112x112x1x1]
%onnx::Conv_891[FLOAT, 112x1x3x3]
%onnx::Conv_894[FLOAT, 112x112x1x1]
%onnx::Conv_897[FLOAT, 112x56x1x1]
%onnx::Conv_900[FLOAT, 112x1x5x5]
%onnx::Conv_903[FLOAT, 112x56x1x1]
%onnx::Conv_906[FLOAT, 672x112x1x1]
%onnx::Conv_907[FLOAT, 672]
%onnx::Conv_909[FLOAT, 672x1x5x5]
%onnx::Conv_912[FLOAT, 184x672x1x1]
%onnx::Conv_913[FLOAT, 184]
%onnx::Conv_915[FLOAT, 184x184x1x1]
%onnx::Conv_918[FLOAT, 184x1x3x3]
%onnx::Conv_921[FLOAT, 184x184x1x1]
%onnx::Conv_924[FLOAT, 552x184x1x1]
%onnx::Conv_925[FLOAT, 552]
%onnx::Conv_927[FLOAT, 552x1x5x5]
%onnx::Conv_930[FLOAT, 184x552x1x1]
%onnx::Conv_933[FLOAT, 184x92x1x1]
%onnx::Conv_936[FLOAT, 184x1x3x3]
%onnx::Conv_939[FLOAT, 184x92x1x1]
%onnx::Conv_942[FLOAT, 1104x184x1x1]
%onnx::Conv_943[FLOAT, 1104]
%onnx::Conv_945[FLOAT, 1104x1x5x5]
%onnx::Conv_948[FLOAT, 352x1104x1x1]
%onnx::Conv_949[FLOAT, 352]
%onnx::Conv_951[FLOAT, 1504x352x1x1]
%onnx::Conv_952[FLOAT, 1504]
) {
%onnx::Conv_946 = Identity(%onnx::Conv_943)
%onnx::Conv_940 = Identity(%onnx::Conv_913)
%onnx::Conv_937 = Identity(%onnx::Conv_913)
%onnx::Conv_934 = Identity(%onnx::Conv_913)
%onnx::Conv_931 = Identity(%onnx::Conv_913)
%onnx::Conv_928 = Identity(%onnx::Conv_925)
%onnx::Conv_922 = Identity(%onnx::Conv_913)
%onnx::Conv_919 = Identity(%onnx::Conv_913)
%onnx::Conv_916 = Identity(%onnx::Conv_913)
%onnx::Conv_910 = Identity(%onnx::Conv_907)
%onnx::Conv_904 = Identity(%onnx::Conv_877)
%onnx::Conv_901 = Identity(%onnx::Conv_877)
%onnx::Conv_898 = Identity(%onnx::Conv_877)
%onnx::Conv_895 = Identity(%onnx::Conv_877)
%onnx::Conv_892 = Identity(%onnx::Conv_877)
%onnx::Conv_889 = Identity(%onnx::Conv_877)
%onnx::Conv_886 = Identity(%onnx::Conv_877)
%onnx::Conv_883 = Identity(%onnx::Conv_880)
%onnx::Conv_874 = Identity(%onnx::Conv_841)
%onnx::Conv_871 = Identity(%onnx::Conv_841)
%onnx::Conv_868 = Identity(%onnx::Conv_841)
%onnx::Conv_865 = Identity(%onnx::Conv_835)
%onnx::Conv_862 = Identity(%onnx::Conv_835)
%onnx::Conv_859 = Identity(%onnx::Conv_841)
%onnx::Conv_856 = Identity(%onnx::Conv_841)
%onnx::Conv_853 = Identity(%onnx::Conv_841)
%onnx::Conv_850 = Identity(%onnx::Conv_841)
%onnx::Conv_847 = Identity(%onnx::Conv_841)
%onnx::Conv_844 = Identity(%onnx::Conv_841)
%onnx::Conv_838 = Identity(%onnx::Conv_835)
%onnx::Conv_832 = Identity(%onnx::Conv_805)
%onnx::Conv_829 = Identity(%onnx::Conv_805)
%onnx::Conv_826 = Identity(%onnx::Conv_805)
%onnx::Conv_823 = Identity(%onnx::Conv_805)
%onnx::Conv_820 = Identity(%onnx::Conv_805)
%onnx::Conv_817 = Identity(%onnx::Conv_805)
%onnx::Conv_814 = Identity(%onnx::Conv_805)
%onnx::Conv_811 = Identity(%onnx::Conv_805)
%onnx::Conv_808 = Identity(%onnx::Conv_805)
%onnx::Conv_802 = Identity(%onnx::Conv_778)
%onnx::Conv_799 = Identity(%onnx::Conv_778)
%onnx::Conv_796 = Identity(%onnx::Conv_778)
%onnx::Conv_793 = Identity(%onnx::Conv_778)
%onnx::Conv_790 = Identity(%onnx::Conv_778)
%onnx::Conv_787 = Identity(%onnx::Conv_778)
%onnx::Conv_784 = Identity(%onnx::Conv_778)
%onnx::Conv_781 = Identity(%onnx::Conv_778)
%onnx::Conv_775 = Identity(%onnx::Conv_763)
%onnx::Conv_772 = Identity(%onnx::Conv_763)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_763)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_759, %onnx::Conv_760)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_882, %onnx::Conv_883)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_885, %onnx::Conv_886)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_888, %onnx::Conv_889)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_891, %onnx::Conv_892)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_894, %onnx::Conv_895)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_897, %onnx::Conv_898)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_900, %onnx::Conv_901)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_903, %onnx::Conv_904)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_906, %onnx::Conv_907)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_909, %onnx::Conv_910)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_912, %onnx::Conv_913)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_915, %onnx::Conv_916)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_918, %onnx::Conv_919)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_921, %onnx::Conv_922)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_924, %onnx::Conv_925)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_927, %onnx::Conv_928)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_930, %onnx::Conv_931)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_933, %onnx::Conv_934)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_936, %onnx::Conv_937)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_939, %onnx::Conv_940)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_942, %onnx::Conv_943)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_945, %onnx::Conv_946)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_948, %onnx::Conv_949)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_951, %onnx::Conv_952)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%757 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %757
}
|
val_accuracy
| 0 | 61,950,592 | 2,074,468 |
{'zcp_synflow': 74.26976639511864, 'zcp_zen': 67.22290802001953, 'zcp_epe_nas': 20.623567866690973, 'zcp_fisher': 0.21343660354614258, 'zcp_flops': 61950592.0, 'zcp_grad_norm': 28.06546974182129, 'zcp_grasp': -0.026615142822265625, 'zcp_jacov': -16.050895513416883, 'zcp_l2_norm': 609.3453979492188, 'zcp_nwot': 209.81642875889628, 'zcp_params': 2074468.0, 'zcp_plain': -0.007858202792704105, 'zcp_snip': 49.92510223388672, 'lat_1080ti_1': 0.7363289524229686, 'lat_1080ti_32': 0.6461253580933087, 'lat_1080ti_64': 0.478357963875997, 'lat_2080ti_1': 0.8056036758452221, 'lat_2080ti_32': 0.7164089349856104, 'lat_2080ti_64': 0.5138177674640225, 'lat_essential_ph_1': 0.3018867924528302, 'lat_eyeriss': 0.38640822987915574, 'lat_fpga': 0.3538823855371615, 'lat_gold_6226': 0.32801243368945604, 'lat_gold_6240': 0.6039251368504469, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.4087853162767861, 'lat_raspi4': 0.46727248742240024, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.18110236220472442, 'lat_silver_4114': 0.609937106001622, 'lat_silver_4210r': 0.650301627163342, 'lat_titan_rtx_1': 0.763659125863419, 'lat_titan_rtx_32': 0.7116513889513635, 'lat_titan_rtx_64': 0.5710683335638707, 'lat_titanx_1': 0.40549498380100407, 'lat_titanx_32': 0.6294026974558204, 'lat_titanx_64': 0.45317508807071516, 'lat_titanxp_1': 0.7147696970510206, 'lat_titanxp_32': 0.671675187608265, 'lat_titanxp_64': 0.5030581380997892}
| |
FBNet_546
|
FBNet
|
546
|
546
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_778[FLOAT, 16x3x3x3]
%onnx::Conv_779[FLOAT, 16]
%onnx::Conv_781[FLOAT, 16x16x1x1]
%onnx::Conv_784[FLOAT, 16x1x5x5]
%onnx::Conv_787[FLOAT, 16x16x1x1]
%onnx::Conv_790[FLOAT, 96x16x1x1]
%onnx::Conv_791[FLOAT, 96]
%onnx::Conv_793[FLOAT, 96x1x5x5]
%onnx::Conv_796[FLOAT, 24x96x1x1]
%onnx::Conv_797[FLOAT, 24]
%onnx::Conv_799[FLOAT, 24x12x1x1]
%onnx::Conv_802[FLOAT, 24x1x5x5]
%onnx::Conv_805[FLOAT, 24x12x1x1]
%onnx::Conv_808[FLOAT, 24x12x1x1]
%onnx::Conv_811[FLOAT, 24x1x3x3]
%onnx::Conv_814[FLOAT, 24x12x1x1]
%onnx::Conv_817[FLOAT, 24x12x1x1]
%onnx::Conv_820[FLOAT, 24x1x5x5]
%onnx::Conv_823[FLOAT, 24x12x1x1]
%onnx::Conv_826[FLOAT, 144x24x1x1]
%onnx::Conv_827[FLOAT, 144]
%onnx::Conv_829[FLOAT, 144x1x5x5]
%onnx::Conv_832[FLOAT, 32x144x1x1]
%onnx::Conv_833[FLOAT, 32]
%onnx::Conv_835[FLOAT, 192x32x1x1]
%onnx::Conv_836[FLOAT, 192]
%onnx::Conv_838[FLOAT, 192x1x5x5]
%onnx::Conv_841[FLOAT, 32x192x1x1]
%onnx::Conv_844[FLOAT, 32x16x1x1]
%onnx::Conv_847[FLOAT, 32x1x5x5]
%onnx::Conv_850[FLOAT, 32x16x1x1]
%onnx::Conv_853[FLOAT, 32x32x1x1]
%onnx::Conv_856[FLOAT, 32x1x3x3]
%onnx::Conv_859[FLOAT, 32x32x1x1]
%onnx::Conv_862[FLOAT, 32x16x1x1]
%onnx::Conv_865[FLOAT, 32x1x5x5]
%onnx::Conv_868[FLOAT, 64x16x1x1]
%onnx::Conv_869[FLOAT, 64]
%onnx::Conv_871[FLOAT, 64x64x1x1]
%onnx::Conv_874[FLOAT, 64x1x5x5]
%onnx::Conv_877[FLOAT, 64x64x1x1]
%onnx::Conv_880[FLOAT, 64x32x1x1]
%onnx::Conv_883[FLOAT, 64x1x5x5]
%onnx::Conv_886[FLOAT, 64x32x1x1]
%onnx::Conv_889[FLOAT, 64x32x1x1]
%onnx::Conv_892[FLOAT, 64x1x5x5]
%onnx::Conv_895[FLOAT, 64x32x1x1]
%onnx::Conv_898[FLOAT, 64x32x1x1]
%onnx::Conv_901[FLOAT, 64x1x5x5]
%onnx::Conv_904[FLOAT, 112x32x1x1]
%onnx::Conv_905[FLOAT, 112]
%onnx::Conv_907[FLOAT, 112x56x1x1]
%onnx::Conv_910[FLOAT, 112x1x3x3]
%onnx::Conv_913[FLOAT, 112x56x1x1]
%onnx::Conv_916[FLOAT, 672x112x1x1]
%onnx::Conv_917[FLOAT, 672]
%onnx::Conv_919[FLOAT, 672x1x5x5]
%onnx::Conv_922[FLOAT, 112x672x1x1]
%onnx::Conv_925[FLOAT, 112x56x1x1]
%onnx::Conv_928[FLOAT, 112x1x5x5]
%onnx::Conv_931[FLOAT, 184x56x1x1]
%onnx::Conv_932[FLOAT, 184]
%onnx::Conv_934[FLOAT, 184x184x1x1]
%onnx::Conv_937[FLOAT, 184x1x5x5]
%onnx::Conv_940[FLOAT, 184x184x1x1]
%onnx::Conv_943[FLOAT, 184x92x1x1]
%onnx::Conv_946[FLOAT, 184x1x5x5]
%onnx::Conv_949[FLOAT, 184x92x1x1]
%onnx::Conv_952[FLOAT, 184x184x1x1]
%onnx::Conv_955[FLOAT, 184x1x5x5]
%onnx::Conv_958[FLOAT, 184x184x1x1]
%onnx::Conv_961[FLOAT, 352x184x1x1]
%onnx::Conv_962[FLOAT, 352]
%onnx::Conv_964[FLOAT, 1504x352x1x1]
%onnx::Conv_965[FLOAT, 1504]
) {
%onnx::Conv_959 = Identity(%onnx::Conv_932)
%onnx::Conv_956 = Identity(%onnx::Conv_932)
%onnx::Conv_953 = Identity(%onnx::Conv_932)
%onnx::Conv_950 = Identity(%onnx::Conv_932)
%onnx::Conv_947 = Identity(%onnx::Conv_932)
%onnx::Conv_944 = Identity(%onnx::Conv_932)
%onnx::Conv_941 = Identity(%onnx::Conv_932)
%onnx::Conv_938 = Identity(%onnx::Conv_932)
%onnx::Conv_935 = Identity(%onnx::Conv_932)
%onnx::Conv_929 = Identity(%onnx::Conv_905)
%onnx::Conv_926 = Identity(%onnx::Conv_905)
%onnx::Conv_923 = Identity(%onnx::Conv_905)
%onnx::Conv_920 = Identity(%onnx::Conv_917)
%onnx::Conv_914 = Identity(%onnx::Conv_905)
%onnx::Conv_911 = Identity(%onnx::Conv_905)
%onnx::Conv_908 = Identity(%onnx::Conv_905)
%onnx::Conv_902 = Identity(%onnx::Conv_869)
%onnx::Conv_899 = Identity(%onnx::Conv_869)
%onnx::Conv_896 = Identity(%onnx::Conv_869)
%onnx::Conv_893 = Identity(%onnx::Conv_869)
%onnx::Conv_890 = Identity(%onnx::Conv_869)
%onnx::Conv_887 = Identity(%onnx::Conv_869)
%onnx::Conv_884 = Identity(%onnx::Conv_869)
%onnx::Conv_881 = Identity(%onnx::Conv_869)
%onnx::Conv_878 = Identity(%onnx::Conv_869)
%onnx::Conv_875 = Identity(%onnx::Conv_869)
%onnx::Conv_872 = Identity(%onnx::Conv_869)
%onnx::Conv_866 = Identity(%onnx::Conv_833)
%onnx::Conv_863 = Identity(%onnx::Conv_833)
%onnx::Conv_860 = Identity(%onnx::Conv_833)
%onnx::Conv_857 = Identity(%onnx::Conv_833)
%onnx::Conv_854 = Identity(%onnx::Conv_833)
%onnx::Conv_851 = Identity(%onnx::Conv_833)
%onnx::Conv_848 = Identity(%onnx::Conv_833)
%onnx::Conv_845 = Identity(%onnx::Conv_833)
%onnx::Conv_842 = Identity(%onnx::Conv_833)
%onnx::Conv_839 = Identity(%onnx::Conv_836)
%onnx::Conv_830 = Identity(%onnx::Conv_827)
%onnx::Conv_824 = Identity(%onnx::Conv_797)
%onnx::Conv_821 = Identity(%onnx::Conv_797)
%onnx::Conv_818 = Identity(%onnx::Conv_797)
%onnx::Conv_815 = Identity(%onnx::Conv_797)
%onnx::Conv_812 = Identity(%onnx::Conv_797)
%onnx::Conv_809 = Identity(%onnx::Conv_797)
%onnx::Conv_806 = Identity(%onnx::Conv_797)
%onnx::Conv_803 = Identity(%onnx::Conv_797)
%onnx::Conv_800 = Identity(%onnx::Conv_797)
%onnx::Conv_794 = Identity(%onnx::Conv_791)
%onnx::Conv_788 = Identity(%onnx::Conv_779)
%onnx::Conv_785 = Identity(%onnx::Conv_779)
%onnx::Conv_782 = Identity(%onnx::Conv_779)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_778, %onnx::Conv_779)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_862, %onnx::Conv_863)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_865, %onnx::Conv_866)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_868, %onnx::Conv_869)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_871, %onnx::Conv_872)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_874, %onnx::Conv_875)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_877, %onnx::Conv_878)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_880, %onnx::Conv_881)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_883, %onnx::Conv_884)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_886, %onnx::Conv_887)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_889, %onnx::Conv_890)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_892, %onnx::Conv_893)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_895, %onnx::Conv_896)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_898, %onnx::Conv_899)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_901, %onnx::Conv_902)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_904, %onnx::Conv_905)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_907, %onnx::Conv_908)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_910, %onnx::Conv_911)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_913, %onnx::Conv_914)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_916, %onnx::Conv_917)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_919, %onnx::Conv_920)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_922, %onnx::Conv_923)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_925, %onnx::Conv_926)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_928, %onnx::Conv_929)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_931, %onnx::Conv_932)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_934, %onnx::Conv_935)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_937, %onnx::Conv_938)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_940, %onnx::Conv_941)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_943, %onnx::Conv_944)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_946, %onnx::Conv_947)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_949, %onnx::Conv_950)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_952, %onnx::Conv_953)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_955, %onnx::Conv_956)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_958, %onnx::Conv_959)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_961, %onnx::Conv_962)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_964, %onnx::Conv_965)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%776 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %776
}
|
val_accuracy
| 0 | 51,556,480 | 1,218,564 |
{'zcp_synflow': 72.049500872149, 'zcp_zen': 61.704444885253906, 'zcp_epe_nas': 29.693679854294388, 'zcp_fisher': 0.08378762006759644, 'zcp_flops': 51556480.0, 'zcp_grad_norm': 24.312580108642578, 'zcp_grasp': -0.05464363098144531, 'zcp_jacov': -16.062489642306538, 'zcp_l2_norm': 494.3289794921875, 'zcp_nwot': 210.11553820349314, 'zcp_params': 1218564.0, 'zcp_plain': 0.0025805309414863586, 'zcp_snip': 33.01844787597656, 'lat_1080ti_1': 0.7168901481198797, 'lat_1080ti_32': 0.6864525958557581, 'lat_1080ti_64': 0.5238592233392539, 'lat_2080ti_1': 0.7674995676792778, 'lat_2080ti_32': 0.6716623634024745, 'lat_2080ti_64': 0.544686225234755, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.2699741440962354, 'lat_fpga': 0.1802016643742764, 'lat_gold_6226': 0.11724592803237836, 'lat_gold_6240': 0.3686858392151831, 'lat_pixel2': 0.15217391304347827, 'lat_pixel3': 0.3576696709561806, 'lat_raspi4': 0.3489859723166374, 'lat_samsung_a50': 0.10526315789473684, 'lat_samsung_s7': 0.14173228346456693, 'lat_silver_4114': 0.4202467254022185, 'lat_silver_4210r': 0.4649379577318172, 'lat_titan_rtx_1': 0.7356697642870376, 'lat_titan_rtx_32': 0.6578754751688903, 'lat_titan_rtx_64': 0.5921104925718434, 'lat_titanx_1': 0.3894693335083631, 'lat_titanx_32': 0.6511752996985677, 'lat_titanx_64': 0.5492539819372458, 'lat_titanxp_1': 0.6992582869956534, 'lat_titanxp_32': 0.6505333131386023, 'lat_titanxp_64': 0.5719332142828257}
| |
FBNet_2989
|
FBNet
|
2989
|
2989
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_697[FLOAT, 16x3x3x3]
%onnx::Conv_698[FLOAT, 16]
%onnx::Conv_700[FLOAT, 96x16x1x1]
%onnx::Conv_701[FLOAT, 96]
%onnx::Conv_703[FLOAT, 96x1x5x5]
%onnx::Conv_706[FLOAT, 16x96x1x1]
%onnx::Conv_709[FLOAT, 96x16x1x1]
%onnx::Conv_712[FLOAT, 96x1x5x5]
%onnx::Conv_715[FLOAT, 24x96x1x1]
%onnx::Conv_716[FLOAT, 24]
%onnx::Conv_718[FLOAT, 24x12x1x1]
%onnx::Conv_721[FLOAT, 24x1x3x3]
%onnx::Conv_724[FLOAT, 24x12x1x1]
%onnx::Conv_727[FLOAT, 72x24x1x1]
%onnx::Conv_728[FLOAT, 72]
%onnx::Conv_730[FLOAT, 72x1x5x5]
%onnx::Conv_733[FLOAT, 24x72x1x1]
%onnx::Conv_736[FLOAT, 32x24x1x1]
%onnx::Conv_737[FLOAT, 32]
%onnx::Conv_739[FLOAT, 96x32x1x1]
%onnx::Conv_742[FLOAT, 96x1x5x5]
%onnx::Conv_745[FLOAT, 32x96x1x1]
%onnx::Conv_748[FLOAT, 32x16x1x1]
%onnx::Conv_751[FLOAT, 32x1x5x5]
%onnx::Conv_754[FLOAT, 32x16x1x1]
%onnx::Conv_757[FLOAT, 32x32x1x1]
%onnx::Conv_760[FLOAT, 32x1x3x3]
%onnx::Conv_763[FLOAT, 64x32x1x1]
%onnx::Conv_764[FLOAT, 64]
%onnx::Conv_766[FLOAT, 64x32x1x1]
%onnx::Conv_769[FLOAT, 64x1x5x5]
%onnx::Conv_772[FLOAT, 64x32x1x1]
%onnx::Conv_775[FLOAT, 64x32x1x1]
%onnx::Conv_778[FLOAT, 64x1x3x3]
%onnx::Conv_781[FLOAT, 64x32x1x1]
%onnx::Conv_784[FLOAT, 384x64x1x1]
%onnx::Conv_785[FLOAT, 384]
%onnx::Conv_787[FLOAT, 384x1x3x3]
%onnx::Conv_790[FLOAT, 64x384x1x1]
%onnx::Conv_793[FLOAT, 384x64x1x1]
%onnx::Conv_796[FLOAT, 384x1x5x5]
%onnx::Conv_799[FLOAT, 112x384x1x1]
%onnx::Conv_800[FLOAT, 112]
%onnx::Conv_802[FLOAT, 112x56x1x1]
%onnx::Conv_805[FLOAT, 112x1x3x3]
%onnx::Conv_808[FLOAT, 112x56x1x1]
%onnx::Conv_811[FLOAT, 112x56x1x1]
%onnx::Conv_814[FLOAT, 112x1x5x5]
%onnx::Conv_817[FLOAT, 112x56x1x1]
%onnx::Conv_820[FLOAT, 112x112x1x1]
%onnx::Conv_823[FLOAT, 112x1x5x5]
%onnx::Conv_826[FLOAT, 112x112x1x1]
%onnx::Conv_829[FLOAT, 112x56x1x1]
%onnx::Conv_832[FLOAT, 112x1x5x5]
%onnx::Conv_835[FLOAT, 184x56x1x1]
%onnx::Conv_836[FLOAT, 184]
%onnx::Conv_838[FLOAT, 184x184x1x1]
%onnx::Conv_841[FLOAT, 184x1x5x5]
%onnx::Conv_844[FLOAT, 184x184x1x1]
%onnx::Conv_847[FLOAT, 184x92x1x1]
%onnx::Conv_850[FLOAT, 184x1x3x3]
%onnx::Conv_853[FLOAT, 184x92x1x1]
%onnx::Conv_856[FLOAT, 184x184x1x1]
%onnx::Conv_859[FLOAT, 184x1x5x5]
%onnx::Conv_862[FLOAT, 184x184x1x1]
%onnx::Conv_865[FLOAT, 184x184x1x1]
%onnx::Conv_868[FLOAT, 184x1x5x5]
%onnx::Conv_871[FLOAT, 352x184x1x1]
%onnx::Conv_872[FLOAT, 352]
%onnx::Conv_874[FLOAT, 1504x352x1x1]
%onnx::Conv_875[FLOAT, 1504]
) {
%onnx::Conv_869 = Identity(%onnx::Conv_836)
%onnx::Conv_866 = Identity(%onnx::Conv_836)
%onnx::Conv_863 = Identity(%onnx::Conv_836)
%onnx::Conv_860 = Identity(%onnx::Conv_836)
%onnx::Conv_857 = Identity(%onnx::Conv_836)
%onnx::Conv_854 = Identity(%onnx::Conv_836)
%onnx::Conv_851 = Identity(%onnx::Conv_836)
%onnx::Conv_848 = Identity(%onnx::Conv_836)
%onnx::Conv_845 = Identity(%onnx::Conv_836)
%onnx::Conv_842 = Identity(%onnx::Conv_836)
%onnx::Conv_839 = Identity(%onnx::Conv_836)
%onnx::Conv_833 = Identity(%onnx::Conv_800)
%onnx::Conv_830 = Identity(%onnx::Conv_800)
%onnx::Conv_827 = Identity(%onnx::Conv_800)
%onnx::Conv_824 = Identity(%onnx::Conv_800)
%onnx::Conv_821 = Identity(%onnx::Conv_800)
%onnx::Conv_818 = Identity(%onnx::Conv_800)
%onnx::Conv_815 = Identity(%onnx::Conv_800)
%onnx::Conv_812 = Identity(%onnx::Conv_800)
%onnx::Conv_809 = Identity(%onnx::Conv_800)
%onnx::Conv_806 = Identity(%onnx::Conv_800)
%onnx::Conv_803 = Identity(%onnx::Conv_800)
%onnx::Conv_797 = Identity(%onnx::Conv_785)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_764)
%onnx::Conv_788 = Identity(%onnx::Conv_785)
%onnx::Conv_782 = Identity(%onnx::Conv_764)
%onnx::Conv_779 = Identity(%onnx::Conv_764)
%onnx::Conv_776 = Identity(%onnx::Conv_764)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_764)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_761 = Identity(%onnx::Conv_737)
%onnx::Conv_758 = Identity(%onnx::Conv_737)
%onnx::Conv_755 = Identity(%onnx::Conv_737)
%onnx::Conv_752 = Identity(%onnx::Conv_737)
%onnx::Conv_749 = Identity(%onnx::Conv_737)
%onnx::Conv_746 = Identity(%onnx::Conv_737)
%onnx::Conv_743 = Identity(%onnx::Conv_701)
%onnx::Conv_740 = Identity(%onnx::Conv_701)
%onnx::Conv_734 = Identity(%onnx::Conv_716)
%onnx::Conv_731 = Identity(%onnx::Conv_728)
%onnx::Conv_725 = Identity(%onnx::Conv_716)
%onnx::Conv_722 = Identity(%onnx::Conv_716)
%onnx::Conv_719 = Identity(%onnx::Conv_716)
%onnx::Conv_713 = Identity(%onnx::Conv_701)
%onnx::Conv_710 = Identity(%onnx::Conv_701)
%onnx::Conv_707 = Identity(%onnx::Conv_698)
%onnx::Conv_704 = Identity(%onnx::Conv_701)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_697, %onnx::Conv_698)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.4/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_862, %onnx::Conv_863)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_865, %onnx::Conv_866)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_868, %onnx::Conv_869)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_871, %onnx::Conv_872)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_874, %onnx::Conv_875)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%695 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %695
}
|
val_accuracy
| 0 | 50,902,912 | 1,229,380 |
{'zcp_synflow': 71.70699097024588, 'zcp_zen': 63.208370208740234, 'zcp_epe_nas': 17.72535093550989, 'zcp_fisher': 0.13038179278373718, 'zcp_flops': 50902912.0, 'zcp_grad_norm': 26.228412628173828, 'zcp_grasp': 0.08196258544921875, 'zcp_jacov': -16.061432330386754, 'zcp_l2_norm': 523.8292846679688, 'zcp_nwot': 209.48636721595247, 'zcp_params': 1229380.0, 'zcp_plain': 0.0007801461033523083, 'zcp_snip': 42.10314178466797, 'lat_1080ti_1': 0.6693414210665644, 'lat_1080ti_32': 0.5834898477082179, 'lat_1080ti_64': 0.4957775766417555, 'lat_2080ti_1': 0.627027492468427, 'lat_2080ti_32': 0.6139499718497835, 'lat_2080ti_64': 0.49921257895911825, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.2989107512882108, 'lat_fpga': 0.21584420981802718, 'lat_gold_6226': 0.112485104287194, 'lat_gold_6240': 0.2779651813764028, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.33648152046768764, 'lat_raspi4': 0.3018848610174463, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.09448818897637795, 'lat_silver_4114': 0.29783739593751274, 'lat_silver_4210r': 0.3612636517795791, 'lat_titan_rtx_1': 0.5920603224989814, 'lat_titan_rtx_32': 0.5900337409315007, 'lat_titan_rtx_64': 0.5420804740604647, 'lat_titanx_1': 0.32568992322396906, 'lat_titanx_32': 0.5679947832871244, 'lat_titanx_64': 0.48730489257169113, 'lat_titanxp_1': 0.5547048825983137, 'lat_titanxp_32': 0.6091523989899575, 'lat_titanxp_64': 0.5129280723098871}
| |
FBNet_4384
|
FBNet
|
4384
|
4384
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_653[FLOAT, 16x3x3x3]
%onnx::Conv_654[FLOAT, 16]
%onnx::Conv_656[FLOAT, 16x16x1x1]
%onnx::Conv_659[FLOAT, 16x1x3x3]
%onnx::Conv_662[FLOAT, 16x16x1x1]
%onnx::Conv_665[FLOAT, 16x8x1x1]
%onnx::Conv_668[FLOAT, 16x1x5x5]
%onnx::Conv_671[FLOAT, 24x8x1x1]
%onnx::Conv_672[FLOAT, 24]
%onnx::Conv_674[FLOAT, 72x24x1x1]
%onnx::Conv_675[FLOAT, 72]
%onnx::Conv_677[FLOAT, 72x1x5x5]
%onnx::Conv_680[FLOAT, 24x72x1x1]
%onnx::Conv_683[FLOAT, 144x24x1x1]
%onnx::Conv_684[FLOAT, 144]
%onnx::Conv_686[FLOAT, 144x1x5x5]
%onnx::Conv_689[FLOAT, 24x144x1x1]
%onnx::Conv_692[FLOAT, 72x24x1x1]
%onnx::Conv_695[FLOAT, 72x1x3x3]
%onnx::Conv_698[FLOAT, 24x72x1x1]
%onnx::Conv_701[FLOAT, 32x24x1x1]
%onnx::Conv_702[FLOAT, 32]
%onnx::Conv_704[FLOAT, 32x16x1x1]
%onnx::Conv_707[FLOAT, 32x1x5x5]
%onnx::Conv_710[FLOAT, 32x16x1x1]
%onnx::Conv_713[FLOAT, 32x32x1x1]
%onnx::Conv_716[FLOAT, 32x1x3x3]
%onnx::Conv_719[FLOAT, 32x32x1x1]
%onnx::Conv_722[FLOAT, 32x16x1x1]
%onnx::Conv_725[FLOAT, 32x1x5x5]
%onnx::Conv_728[FLOAT, 32x16x1x1]
%onnx::Conv_731[FLOAT, 32x16x1x1]
%onnx::Conv_734[FLOAT, 32x1x3x3]
%onnx::Conv_737[FLOAT, 64x16x1x1]
%onnx::Conv_738[FLOAT, 64]
%onnx::Conv_740[FLOAT, 64x64x1x1]
%onnx::Conv_743[FLOAT, 64x1x3x3]
%onnx::Conv_746[FLOAT, 64x64x1x1]
%onnx::Conv_749[FLOAT, 64x32x1x1]
%onnx::Conv_752[FLOAT, 64x1x5x5]
%onnx::Conv_755[FLOAT, 64x32x1x1]
%onnx::Conv_758[FLOAT, 192x64x1x1]
%onnx::Conv_759[FLOAT, 192]
%onnx::Conv_761[FLOAT, 192x1x3x3]
%onnx::Conv_764[FLOAT, 64x192x1x1]
%onnx::Conv_767[FLOAT, 112x64x1x1]
%onnx::Conv_768[FLOAT, 112]
%onnx::Conv_770[FLOAT, 112x112x1x1]
%onnx::Conv_773[FLOAT, 112x1x5x5]
%onnx::Conv_776[FLOAT, 112x112x1x1]
%onnx::Conv_779[FLOAT, 112x56x1x1]
%onnx::Conv_782[FLOAT, 112x1x5x5]
%onnx::Conv_785[FLOAT, 112x56x1x1]
%onnx::Conv_788[FLOAT, 336x112x1x1]
%onnx::Conv_789[FLOAT, 336]
%onnx::Conv_791[FLOAT, 336x1x3x3]
%onnx::Conv_794[FLOAT, 112x336x1x1]
%onnx::Conv_797[FLOAT, 112x112x1x1]
%onnx::Conv_800[FLOAT, 112x1x3x3]
%onnx::Conv_803[FLOAT, 184x112x1x1]
%onnx::Conv_804[FLOAT, 184]
%onnx::Conv_806[FLOAT, 184x184x1x1]
%onnx::Conv_809[FLOAT, 184x1x3x3]
%onnx::Conv_812[FLOAT, 184x184x1x1]
%onnx::Conv_815[FLOAT, 1104x184x1x1]
%onnx::Conv_816[FLOAT, 1104]
%onnx::Conv_818[FLOAT, 1104x1x3x3]
%onnx::Conv_821[FLOAT, 184x1104x1x1]
%onnx::Conv_824[FLOAT, 352x184x1x1]
%onnx::Conv_825[FLOAT, 352]
%onnx::Conv_827[FLOAT, 1504x352x1x1]
%onnx::Conv_828[FLOAT, 1504]
) {
%onnx::Conv_822 = Identity(%onnx::Conv_804)
%onnx::Conv_819 = Identity(%onnx::Conv_816)
%onnx::Conv_813 = Identity(%onnx::Conv_804)
%onnx::Conv_810 = Identity(%onnx::Conv_804)
%onnx::Conv_807 = Identity(%onnx::Conv_804)
%onnx::Conv_801 = Identity(%onnx::Conv_768)
%onnx::Conv_798 = Identity(%onnx::Conv_768)
%onnx::Conv_795 = Identity(%onnx::Conv_768)
%onnx::Conv_792 = Identity(%onnx::Conv_789)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_768)
%onnx::Conv_780 = Identity(%onnx::Conv_768)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_768)
%onnx::Conv_771 = Identity(%onnx::Conv_768)
%onnx::Conv_765 = Identity(%onnx::Conv_738)
%onnx::Conv_762 = Identity(%onnx::Conv_759)
%onnx::Conv_756 = Identity(%onnx::Conv_738)
%onnx::Conv_753 = Identity(%onnx::Conv_738)
%onnx::Conv_750 = Identity(%onnx::Conv_738)
%onnx::Conv_747 = Identity(%onnx::Conv_738)
%onnx::Conv_744 = Identity(%onnx::Conv_738)
%onnx::Conv_741 = Identity(%onnx::Conv_738)
%onnx::Conv_735 = Identity(%onnx::Conv_702)
%onnx::Conv_732 = Identity(%onnx::Conv_702)
%onnx::Conv_729 = Identity(%onnx::Conv_702)
%onnx::Conv_726 = Identity(%onnx::Conv_702)
%onnx::Conv_723 = Identity(%onnx::Conv_702)
%onnx::Conv_720 = Identity(%onnx::Conv_702)
%onnx::Conv_717 = Identity(%onnx::Conv_702)
%onnx::Conv_714 = Identity(%onnx::Conv_702)
%onnx::Conv_711 = Identity(%onnx::Conv_702)
%onnx::Conv_708 = Identity(%onnx::Conv_702)
%onnx::Conv_705 = Identity(%onnx::Conv_702)
%onnx::Conv_699 = Identity(%onnx::Conv_672)
%onnx::Conv_696 = Identity(%onnx::Conv_675)
%onnx::Conv_693 = Identity(%onnx::Conv_675)
%onnx::Conv_690 = Identity(%onnx::Conv_672)
%onnx::Conv_687 = Identity(%onnx::Conv_684)
%onnx::Conv_681 = Identity(%onnx::Conv_672)
%onnx::Conv_678 = Identity(%onnx::Conv_675)
%onnx::Conv_669 = Identity(%onnx::Conv_654)
%onnx::Conv_666 = Identity(%onnx::Conv_654)
%onnx::Conv_663 = Identity(%onnx::Conv_654)
%onnx::Conv_660 = Identity(%onnx::Conv_654)
%onnx::Conv_657 = Identity(%onnx::Conv_654)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_653, %onnx::Conv_654)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.4/Add_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%651 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %651
}
|
val_accuracy
| 0 | 56,332,672 | 1,481,772 |
{'zcp_synflow': 73.09383755636156, 'zcp_zen': 59.81539535522461, 'zcp_epe_nas': 17.527095286074303, 'zcp_fisher': 0.06531563401222229, 'zcp_flops': 56332672.0, 'zcp_grad_norm': 19.767742156982422, 'zcp_grasp': -0.028875350952148438, 'zcp_jacov': -16.057996338026623, 'zcp_l2_norm': 519.0757446289062, 'zcp_nwot': 210.78938601541066, 'zcp_params': 1481772.0, 'zcp_plain': -0.0019710715860128403, 'zcp_snip': 29.871259689331055, 'lat_1080ti_1': 0.5457327454936964, 'lat_1080ti_32': 0.6386818162589222, 'lat_1080ti_64': 0.4825904490192845, 'lat_2080ti_1': 0.5919187725786728, 'lat_2080ti_32': 0.5957195513983291, 'lat_2080ti_64': 0.5192371714547994, 'lat_essential_ph_1': 0.20754716981132076, 'lat_eyeriss': 0.31194872829296005, 'lat_fpga': 0.2851601539694634, 'lat_gold_6226': 0.15859147149832437, 'lat_gold_6240': 0.23279079046718407, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.35301336523507143, 'lat_raspi4': 0.3685985500386761, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.08661417322834646, 'lat_silver_4114': 0.2392220258180565, 'lat_silver_4210r': 0.2493173659045874, 'lat_titan_rtx_1': 0.5369767257164554, 'lat_titan_rtx_32': 0.564078639217786, 'lat_titan_rtx_64': 0.5348211458022154, 'lat_titanx_1': 0.2820536281486439, 'lat_titanx_32': 0.5617058228085927, 'lat_titanx_64': 0.4724561488840589, 'lat_titanxp_1': 0.49437634009642983, 'lat_titanxp_32': 0.5682602647887893, 'lat_titanxp_64': 0.5075069637920623}
| |
FBNet_3027
|
FBNet
|
3027
|
3027
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_576[FLOAT, 16x3x3x3]
%onnx::Conv_577[FLOAT, 16]
%onnx::Conv_579[FLOAT, 16x16x1x1]
%onnx::Conv_582[FLOAT, 16x1x3x3]
%onnx::Conv_585[FLOAT, 16x16x1x1]
%onnx::Conv_588[FLOAT, 96x16x1x1]
%onnx::Conv_589[FLOAT, 96]
%onnx::Conv_591[FLOAT, 96x1x3x3]
%onnx::Conv_594[FLOAT, 24x96x1x1]
%onnx::Conv_595[FLOAT, 24]
%onnx::Conv_597[FLOAT, 72x24x1x1]
%onnx::Conv_598[FLOAT, 72]
%onnx::Conv_600[FLOAT, 72x1x5x5]
%onnx::Conv_603[FLOAT, 24x72x1x1]
%onnx::Conv_606[FLOAT, 24x24x1x1]
%onnx::Conv_609[FLOAT, 24x1x3x3]
%onnx::Conv_612[FLOAT, 24x24x1x1]
%onnx::Conv_615[FLOAT, 24x24x1x1]
%onnx::Conv_618[FLOAT, 24x1x3x3]
%onnx::Conv_621[FLOAT, 32x24x1x1]
%onnx::Conv_622[FLOAT, 32]
%onnx::Conv_624[FLOAT, 32x32x1x1]
%onnx::Conv_627[FLOAT, 32x1x3x3]
%onnx::Conv_630[FLOAT, 32x32x1x1]
%onnx::Conv_633[FLOAT, 32x32x1x1]
%onnx::Conv_636[FLOAT, 32x1x5x5]
%onnx::Conv_639[FLOAT, 32x32x1x1]
%onnx::Conv_642[FLOAT, 192x32x1x1]
%onnx::Conv_643[FLOAT, 192]
%onnx::Conv_645[FLOAT, 192x1x3x3]
%onnx::Conv_648[FLOAT, 64x192x1x1]
%onnx::Conv_649[FLOAT, 64]
%onnx::Conv_651[FLOAT, 192x64x1x1]
%onnx::Conv_654[FLOAT, 192x1x3x3]
%onnx::Conv_657[FLOAT, 64x192x1x1]
%onnx::Conv_660[FLOAT, 64x64x1x1]
%onnx::Conv_663[FLOAT, 64x1x3x3]
%onnx::Conv_666[FLOAT, 64x64x1x1]
%onnx::Conv_669[FLOAT, 64x64x1x1]
%onnx::Conv_672[FLOAT, 64x1x5x5]
%onnx::Conv_675[FLOAT, 112x64x1x1]
%onnx::Conv_676[FLOAT, 112]
%onnx::Conv_678[FLOAT, 336x112x1x1]
%onnx::Conv_679[FLOAT, 336]
%onnx::Conv_681[FLOAT, 336x1x3x3]
%onnx::Conv_684[FLOAT, 112x336x1x1]
%onnx::Conv_687[FLOAT, 672x112x1x1]
%onnx::Conv_688[FLOAT, 672]
%onnx::Conv_690[FLOAT, 672x1x5x5]
%onnx::Conv_693[FLOAT, 112x672x1x1]
%onnx::Conv_696[FLOAT, 112x56x1x1]
%onnx::Conv_699[FLOAT, 112x1x3x3]
%onnx::Conv_702[FLOAT, 112x56x1x1]
%onnx::Conv_705[FLOAT, 336x112x1x1]
%onnx::Conv_708[FLOAT, 336x1x3x3]
%onnx::Conv_711[FLOAT, 184x336x1x1]
%onnx::Conv_712[FLOAT, 184]
%onnx::Conv_714[FLOAT, 1104x184x1x1]
%onnx::Conv_715[FLOAT, 1104]
%onnx::Conv_717[FLOAT, 1104x1x5x5]
%onnx::Conv_720[FLOAT, 184x1104x1x1]
%onnx::Conv_723[FLOAT, 184x184x1x1]
%onnx::Conv_726[FLOAT, 184x1x5x5]
%onnx::Conv_729[FLOAT, 184x184x1x1]
%onnx::Conv_732[FLOAT, 184x92x1x1]
%onnx::Conv_735[FLOAT, 184x1x5x5]
%onnx::Conv_738[FLOAT, 184x92x1x1]
%onnx::Conv_741[FLOAT, 552x184x1x1]
%onnx::Conv_742[FLOAT, 552]
%onnx::Conv_744[FLOAT, 552x1x5x5]
%onnx::Conv_747[FLOAT, 352x552x1x1]
%onnx::Conv_748[FLOAT, 352]
%onnx::Conv_750[FLOAT, 1504x352x1x1]
%onnx::Conv_751[FLOAT, 1504]
) {
%onnx::Conv_745 = Identity(%onnx::Conv_742)
%onnx::Conv_739 = Identity(%onnx::Conv_712)
%onnx::Conv_736 = Identity(%onnx::Conv_712)
%onnx::Conv_733 = Identity(%onnx::Conv_712)
%onnx::Conv_730 = Identity(%onnx::Conv_712)
%onnx::Conv_727 = Identity(%onnx::Conv_712)
%onnx::Conv_724 = Identity(%onnx::Conv_712)
%onnx::Conv_721 = Identity(%onnx::Conv_712)
%onnx::Conv_718 = Identity(%onnx::Conv_715)
%onnx::Conv_709 = Identity(%onnx::Conv_679)
%onnx::Conv_706 = Identity(%onnx::Conv_679)
%onnx::Conv_703 = Identity(%onnx::Conv_676)
%onnx::Conv_700 = Identity(%onnx::Conv_676)
%onnx::Conv_697 = Identity(%onnx::Conv_676)
%onnx::Conv_694 = Identity(%onnx::Conv_676)
%onnx::Conv_691 = Identity(%onnx::Conv_688)
%onnx::Conv_685 = Identity(%onnx::Conv_676)
%onnx::Conv_682 = Identity(%onnx::Conv_679)
%onnx::Conv_673 = Identity(%onnx::Conv_649)
%onnx::Conv_670 = Identity(%onnx::Conv_649)
%onnx::Conv_667 = Identity(%onnx::Conv_649)
%onnx::Conv_664 = Identity(%onnx::Conv_649)
%onnx::Conv_661 = Identity(%onnx::Conv_649)
%onnx::Conv_658 = Identity(%onnx::Conv_649)
%onnx::Conv_655 = Identity(%onnx::Conv_643)
%onnx::Conv_652 = Identity(%onnx::Conv_643)
%onnx::Conv_646 = Identity(%onnx::Conv_643)
%onnx::Conv_640 = Identity(%onnx::Conv_622)
%onnx::Conv_637 = Identity(%onnx::Conv_622)
%onnx::Conv_634 = Identity(%onnx::Conv_622)
%onnx::Conv_631 = Identity(%onnx::Conv_622)
%onnx::Conv_628 = Identity(%onnx::Conv_622)
%onnx::Conv_625 = Identity(%onnx::Conv_622)
%onnx::Conv_619 = Identity(%onnx::Conv_595)
%onnx::Conv_616 = Identity(%onnx::Conv_595)
%onnx::Conv_613 = Identity(%onnx::Conv_595)
%onnx::Conv_610 = Identity(%onnx::Conv_595)
%onnx::Conv_607 = Identity(%onnx::Conv_595)
%onnx::Conv_604 = Identity(%onnx::Conv_595)
%onnx::Conv_601 = Identity(%onnx::Conv_598)
%onnx::Conv_592 = Identity(%onnx::Conv_589)
%onnx::Conv_586 = Identity(%onnx::Conv_577)
%onnx::Conv_583 = Identity(%onnx::Conv_577)
%onnx::Conv_580 = Identity(%onnx::Conv_577)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_576, %onnx::Conv_577)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_579, %onnx::Conv_580)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_582, %onnx::Conv_583)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_585, %onnx::Conv_586)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_588, %onnx::Conv_589)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_591, %onnx::Conv_592)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_594, %onnx::Conv_595)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_597, %onnx::Conv_598)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_600, %onnx::Conv_601)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_603, %onnx::Conv_604)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_606, %onnx::Conv_607)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_609, %onnx::Conv_610)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_612, %onnx::Conv_613)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_615, %onnx::Conv_616)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%574 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %574
}
|
val_accuracy
| 0 | 65,266,304 | 2,007,140 |
{'zcp_synflow': 75.02481123629833, 'zcp_zen': 65.25361633300781, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.06665295362472534, 'zcp_flops': 65266304.0, 'zcp_grad_norm': 19.282649993896484, 'zcp_grasp': -0.057211875915527344, 'zcp_jacov': -16.049610187783554, 'zcp_l2_norm': 608.7345581054688, 'zcp_nwot': 209.34543279987832, 'zcp_params': 2007140.0, 'zcp_plain': 0.0008698090678080916, 'zcp_snip': 36.83035659790039, 'lat_1080ti_1': 0.4531367897112039, 'lat_1080ti_32': 0.39699390833663045, 'lat_1080ti_64': 0.2950132612060024, 'lat_2080ti_1': 0.4594507598864291, 'lat_2080ti_32': 0.37343216523160927, 'lat_2080ti_64': 0.3169513461711893, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.36429318027616303, 'lat_fpga': 0.43077806034098887, 'lat_gold_6226': 0.3396261962517624, 'lat_gold_6240': 0.4288389933971198, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.34783363381217636, 'lat_raspi4': 0.4092197528829237, 'lat_samsung_a50': 0.22105263157894736, 'lat_samsung_s7': 0.14173228346456693, 'lat_silver_4114': 0.5258810999686425, 'lat_silver_4210r': 0.44178948508243654, 'lat_titan_rtx_1': 0.42904413939963065, 'lat_titan_rtx_32': 0.36322677248521, 'lat_titan_rtx_64': 0.3249386319077661, 'lat_titanx_1': 0.22886739778627688, 'lat_titanx_32': 0.34271635655649, 'lat_titanx_64': 0.29203283218065557, 'lat_titanxp_1': 0.4039470465497667, 'lat_titanxp_32': 0.37420012500543326, 'lat_titanxp_64': 0.3140747381237042}
| |
FBNet_3257
|
FBNet
|
3257
|
3257
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_688[FLOAT, 16x3x3x3]
%onnx::Conv_689[FLOAT, 16]
%onnx::Conv_691[FLOAT, 48x16x1x1]
%onnx::Conv_692[FLOAT, 48]
%onnx::Conv_694[FLOAT, 48x1x5x5]
%onnx::Conv_697[FLOAT, 16x48x1x1]
%onnx::Conv_700[FLOAT, 16x8x1x1]
%onnx::Conv_703[FLOAT, 16x1x3x3]
%onnx::Conv_706[FLOAT, 24x8x1x1]
%onnx::Conv_707[FLOAT, 24]
%onnx::Conv_709[FLOAT, 144x24x1x1]
%onnx::Conv_710[FLOAT, 144]
%onnx::Conv_712[FLOAT, 144x1x5x5]
%onnx::Conv_715[FLOAT, 24x144x1x1]
%onnx::Conv_718[FLOAT, 24x24x1x1]
%onnx::Conv_721[FLOAT, 24x1x3x3]
%onnx::Conv_724[FLOAT, 24x24x1x1]
%onnx::Conv_727[FLOAT, 144x24x1x1]
%onnx::Conv_730[FLOAT, 144x1x3x3]
%onnx::Conv_733[FLOAT, 24x144x1x1]
%onnx::Conv_736[FLOAT, 24x12x1x1]
%onnx::Conv_739[FLOAT, 24x1x5x5]
%onnx::Conv_742[FLOAT, 32x12x1x1]
%onnx::Conv_743[FLOAT, 32]
%onnx::Conv_745[FLOAT, 96x32x1x1]
%onnx::Conv_746[FLOAT, 96]
%onnx::Conv_748[FLOAT, 96x1x5x5]
%onnx::Conv_751[FLOAT, 32x96x1x1]
%onnx::Conv_754[FLOAT, 192x32x1x1]
%onnx::Conv_755[FLOAT, 192]
%onnx::Conv_757[FLOAT, 192x1x3x3]
%onnx::Conv_760[FLOAT, 32x192x1x1]
%onnx::Conv_763[FLOAT, 192x32x1x1]
%onnx::Conv_766[FLOAT, 192x1x5x5]
%onnx::Conv_769[FLOAT, 64x192x1x1]
%onnx::Conv_770[FLOAT, 64]
%onnx::Conv_772[FLOAT, 64x64x1x1]
%onnx::Conv_775[FLOAT, 64x1x5x5]
%onnx::Conv_778[FLOAT, 64x64x1x1]
%onnx::Conv_781[FLOAT, 64x64x1x1]
%onnx::Conv_784[FLOAT, 64x1x3x3]
%onnx::Conv_787[FLOAT, 64x64x1x1]
%onnx::Conv_790[FLOAT, 64x64x1x1]
%onnx::Conv_793[FLOAT, 64x1x3x3]
%onnx::Conv_796[FLOAT, 64x64x1x1]
%onnx::Conv_799[FLOAT, 64x32x1x1]
%onnx::Conv_802[FLOAT, 64x1x3x3]
%onnx::Conv_805[FLOAT, 112x32x1x1]
%onnx::Conv_806[FLOAT, 112]
%onnx::Conv_808[FLOAT, 336x112x1x1]
%onnx::Conv_809[FLOAT, 336]
%onnx::Conv_811[FLOAT, 336x1x3x3]
%onnx::Conv_814[FLOAT, 112x336x1x1]
%onnx::Conv_817[FLOAT, 112x112x1x1]
%onnx::Conv_820[FLOAT, 112x1x3x3]
%onnx::Conv_823[FLOAT, 112x112x1x1]
%onnx::Conv_826[FLOAT, 112x56x1x1]
%onnx::Conv_829[FLOAT, 112x1x3x3]
%onnx::Conv_832[FLOAT, 112x56x1x1]
%onnx::Conv_835[FLOAT, 184x112x1x1]
%onnx::Conv_836[FLOAT, 184]
%onnx::Conv_838[FLOAT, 184x92x1x1]
%onnx::Conv_841[FLOAT, 184x1x3x3]
%onnx::Conv_844[FLOAT, 184x92x1x1]
%onnx::Conv_847[FLOAT, 552x184x1x1]
%onnx::Conv_848[FLOAT, 552]
%onnx::Conv_850[FLOAT, 552x1x3x3]
%onnx::Conv_853[FLOAT, 184x552x1x1]
%onnx::Conv_856[FLOAT, 552x184x1x1]
%onnx::Conv_859[FLOAT, 552x1x3x3]
%onnx::Conv_862[FLOAT, 184x552x1x1]
%onnx::Conv_865[FLOAT, 184x92x1x1]
%onnx::Conv_868[FLOAT, 184x1x5x5]
%onnx::Conv_871[FLOAT, 352x92x1x1]
%onnx::Conv_872[FLOAT, 352]
%onnx::Conv_874[FLOAT, 1504x352x1x1]
%onnx::Conv_875[FLOAT, 1504]
) {
%onnx::Conv_869 = Identity(%onnx::Conv_836)
%onnx::Conv_866 = Identity(%onnx::Conv_836)
%onnx::Conv_863 = Identity(%onnx::Conv_836)
%onnx::Conv_860 = Identity(%onnx::Conv_848)
%onnx::Conv_857 = Identity(%onnx::Conv_848)
%onnx::Conv_854 = Identity(%onnx::Conv_836)
%onnx::Conv_851 = Identity(%onnx::Conv_848)
%onnx::Conv_845 = Identity(%onnx::Conv_836)
%onnx::Conv_842 = Identity(%onnx::Conv_836)
%onnx::Conv_839 = Identity(%onnx::Conv_836)
%onnx::Conv_833 = Identity(%onnx::Conv_806)
%onnx::Conv_830 = Identity(%onnx::Conv_806)
%onnx::Conv_827 = Identity(%onnx::Conv_806)
%onnx::Conv_824 = Identity(%onnx::Conv_806)
%onnx::Conv_821 = Identity(%onnx::Conv_806)
%onnx::Conv_818 = Identity(%onnx::Conv_806)
%onnx::Conv_815 = Identity(%onnx::Conv_806)
%onnx::Conv_812 = Identity(%onnx::Conv_809)
%onnx::Conv_803 = Identity(%onnx::Conv_770)
%onnx::Conv_800 = Identity(%onnx::Conv_770)
%onnx::Conv_797 = Identity(%onnx::Conv_770)
%onnx::Conv_794 = Identity(%onnx::Conv_770)
%onnx::Conv_791 = Identity(%onnx::Conv_770)
%onnx::Conv_788 = Identity(%onnx::Conv_770)
%onnx::Conv_785 = Identity(%onnx::Conv_770)
%onnx::Conv_782 = Identity(%onnx::Conv_770)
%onnx::Conv_779 = Identity(%onnx::Conv_770)
%onnx::Conv_776 = Identity(%onnx::Conv_770)
%onnx::Conv_773 = Identity(%onnx::Conv_770)
%onnx::Conv_767 = Identity(%onnx::Conv_755)
%onnx::Conv_764 = Identity(%onnx::Conv_755)
%onnx::Conv_761 = Identity(%onnx::Conv_743)
%onnx::Conv_758 = Identity(%onnx::Conv_755)
%onnx::Conv_752 = Identity(%onnx::Conv_743)
%onnx::Conv_749 = Identity(%onnx::Conv_746)
%onnx::Conv_740 = Identity(%onnx::Conv_707)
%onnx::Conv_737 = Identity(%onnx::Conv_707)
%onnx::Conv_734 = Identity(%onnx::Conv_707)
%onnx::Conv_731 = Identity(%onnx::Conv_710)
%onnx::Conv_728 = Identity(%onnx::Conv_710)
%onnx::Conv_725 = Identity(%onnx::Conv_707)
%onnx::Conv_722 = Identity(%onnx::Conv_707)
%onnx::Conv_719 = Identity(%onnx::Conv_707)
%onnx::Conv_716 = Identity(%onnx::Conv_707)
%onnx::Conv_713 = Identity(%onnx::Conv_710)
%onnx::Conv_704 = Identity(%onnx::Conv_689)
%onnx::Conv_701 = Identity(%onnx::Conv_689)
%onnx::Conv_698 = Identity(%onnx::Conv_689)
%onnx::Conv_695 = Identity(%onnx::Conv_692)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_688, %onnx::Conv_689)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_862, %onnx::Conv_863)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_865, %onnx::Conv_866)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_868, %onnx::Conv_869)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_871, %onnx::Conv_872)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_874, %onnx::Conv_875)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%686 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %686
}
|
val_accuracy
| 0 | 63,440,128 | 1,448,116 |
{'zcp_synflow': 75.74558689872217, 'zcp_zen': 65.7975082397461, 'zcp_epe_nas': 19.971629817208036, 'zcp_fisher': 0.20391465723514557, 'zcp_flops': 63440128.0, 'zcp_grad_norm': 27.347721099853516, 'zcp_grasp': -0.3170127868652344, 'zcp_jacov': -16.04718261507313, 'zcp_l2_norm': 578.3363037109375, 'zcp_nwot': 214.6753325413459, 'zcp_params': 1448116.0, 'zcp_plain': 0.003302917117252946, 'zcp_snip': 46.11545181274414, 'lat_1080ti_1': 0.6179531173001636, 'lat_1080ti_32': 0.7406024836255123, 'lat_1080ti_64': 0.6040196431479482, 'lat_2080ti_1': 0.7068271971001849, 'lat_2080ti_32': 0.7293357851638166, 'lat_2080ti_64': 0.6366627803084951, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.4217171254103022, 'lat_fpga': 0.39628335922579877, 'lat_gold_6226': 0.22206992073980836, 'lat_gold_6240': 0.41119853708742793, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.4339845238241879, 'lat_raspi4': 0.43508153824755913, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.13385826771653545, 'lat_silver_4114': 0.4573368776771735, 'lat_silver_4210r': 0.5350664900900759, 'lat_titan_rtx_1': 0.6482293767181502, 'lat_titan_rtx_32': 0.6868594950546226, 'lat_titan_rtx_64': 0.6716513698408708, 'lat_titanx_1': 0.34098122035633505, 'lat_titanx_32': 0.6958175977006128, 'lat_titanx_64': 0.5887683864448297, 'lat_titanxp_1': 0.6017244054257371, 'lat_titanxp_32': 0.697161367094077, 'lat_titanxp_64': 0.6114594486085486}
| |
FBNet_1846
|
FBNet
|
1846
|
1846
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_644[FLOAT, 16x3x3x3]
%onnx::Conv_645[FLOAT, 16]
%onnx::Conv_647[FLOAT, 16x8x1x1]
%onnx::Conv_650[FLOAT, 16x1x5x5]
%onnx::Conv_653[FLOAT, 16x8x1x1]
%onnx::Conv_656[FLOAT, 96x16x1x1]
%onnx::Conv_657[FLOAT, 96]
%onnx::Conv_659[FLOAT, 96x1x3x3]
%onnx::Conv_662[FLOAT, 24x96x1x1]
%onnx::Conv_663[FLOAT, 24]
%onnx::Conv_665[FLOAT, 24x12x1x1]
%onnx::Conv_668[FLOAT, 24x1x3x3]
%onnx::Conv_671[FLOAT, 24x12x1x1]
%onnx::Conv_674[FLOAT, 72x24x1x1]
%onnx::Conv_675[FLOAT, 72]
%onnx::Conv_677[FLOAT, 72x1x3x3]
%onnx::Conv_680[FLOAT, 24x72x1x1]
%onnx::Conv_683[FLOAT, 32x24x1x1]
%onnx::Conv_684[FLOAT, 32]
%onnx::Conv_686[FLOAT, 32x32x1x1]
%onnx::Conv_689[FLOAT, 32x1x5x5]
%onnx::Conv_692[FLOAT, 32x32x1x1]
%onnx::Conv_695[FLOAT, 96x32x1x1]
%onnx::Conv_698[FLOAT, 96x1x3x3]
%onnx::Conv_701[FLOAT, 32x96x1x1]
%onnx::Conv_704[FLOAT, 192x32x1x1]
%onnx::Conv_705[FLOAT, 192]
%onnx::Conv_707[FLOAT, 192x1x3x3]
%onnx::Conv_710[FLOAT, 32x192x1x1]
%onnx::Conv_713[FLOAT, 64x32x1x1]
%onnx::Conv_714[FLOAT, 64]
%onnx::Conv_716[FLOAT, 64x32x1x1]
%onnx::Conv_719[FLOAT, 64x1x3x3]
%onnx::Conv_722[FLOAT, 64x32x1x1]
%onnx::Conv_725[FLOAT, 64x64x1x1]
%onnx::Conv_728[FLOAT, 64x1x3x3]
%onnx::Conv_731[FLOAT, 64x64x1x1]
%onnx::Conv_734[FLOAT, 64x32x1x1]
%onnx::Conv_737[FLOAT, 64x1x5x5]
%onnx::Conv_740[FLOAT, 64x32x1x1]
%onnx::Conv_743[FLOAT, 192x64x1x1]
%onnx::Conv_746[FLOAT, 192x1x3x3]
%onnx::Conv_749[FLOAT, 112x192x1x1]
%onnx::Conv_750[FLOAT, 112]
%onnx::Conv_752[FLOAT, 336x112x1x1]
%onnx::Conv_753[FLOAT, 336]
%onnx::Conv_755[FLOAT, 336x1x3x3]
%onnx::Conv_758[FLOAT, 112x336x1x1]
%onnx::Conv_761[FLOAT, 112x56x1x1]
%onnx::Conv_764[FLOAT, 112x1x5x5]
%onnx::Conv_767[FLOAT, 112x56x1x1]
%onnx::Conv_770[FLOAT, 672x112x1x1]
%onnx::Conv_771[FLOAT, 672]
%onnx::Conv_773[FLOAT, 672x1x5x5]
%onnx::Conv_776[FLOAT, 184x672x1x1]
%onnx::Conv_777[FLOAT, 184]
%onnx::Conv_779[FLOAT, 1104x184x1x1]
%onnx::Conv_780[FLOAT, 1104]
%onnx::Conv_782[FLOAT, 1104x1x5x5]
%onnx::Conv_785[FLOAT, 184x1104x1x1]
%onnx::Conv_788[FLOAT, 184x184x1x1]
%onnx::Conv_791[FLOAT, 184x1x3x3]
%onnx::Conv_794[FLOAT, 184x184x1x1]
%onnx::Conv_797[FLOAT, 184x92x1x1]
%onnx::Conv_800[FLOAT, 184x1x3x3]
%onnx::Conv_803[FLOAT, 184x92x1x1]
%onnx::Conv_806[FLOAT, 552x184x1x1]
%onnx::Conv_807[FLOAT, 552]
%onnx::Conv_809[FLOAT, 552x1x5x5]
%onnx::Conv_812[FLOAT, 352x552x1x1]
%onnx::Conv_813[FLOAT, 352]
%onnx::Conv_815[FLOAT, 1504x352x1x1]
%onnx::Conv_816[FLOAT, 1504]
) {
%onnx::Conv_810 = Identity(%onnx::Conv_807)
%onnx::Conv_804 = Identity(%onnx::Conv_777)
%onnx::Conv_801 = Identity(%onnx::Conv_777)
%onnx::Conv_798 = Identity(%onnx::Conv_777)
%onnx::Conv_795 = Identity(%onnx::Conv_777)
%onnx::Conv_792 = Identity(%onnx::Conv_777)
%onnx::Conv_789 = Identity(%onnx::Conv_777)
%onnx::Conv_786 = Identity(%onnx::Conv_777)
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_768 = Identity(%onnx::Conv_750)
%onnx::Conv_765 = Identity(%onnx::Conv_750)
%onnx::Conv_762 = Identity(%onnx::Conv_750)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_747 = Identity(%onnx::Conv_705)
%onnx::Conv_744 = Identity(%onnx::Conv_705)
%onnx::Conv_741 = Identity(%onnx::Conv_714)
%onnx::Conv_738 = Identity(%onnx::Conv_714)
%onnx::Conv_735 = Identity(%onnx::Conv_714)
%onnx::Conv_732 = Identity(%onnx::Conv_714)
%onnx::Conv_729 = Identity(%onnx::Conv_714)
%onnx::Conv_726 = Identity(%onnx::Conv_714)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_714)
%onnx::Conv_717 = Identity(%onnx::Conv_714)
%onnx::Conv_711 = Identity(%onnx::Conv_684)
%onnx::Conv_708 = Identity(%onnx::Conv_705)
%onnx::Conv_702 = Identity(%onnx::Conv_684)
%onnx::Conv_699 = Identity(%onnx::Conv_657)
%onnx::Conv_696 = Identity(%onnx::Conv_657)
%onnx::Conv_693 = Identity(%onnx::Conv_684)
%onnx::Conv_690 = Identity(%onnx::Conv_684)
%onnx::Conv_687 = Identity(%onnx::Conv_684)
%onnx::Conv_681 = Identity(%onnx::Conv_663)
%onnx::Conv_678 = Identity(%onnx::Conv_675)
%onnx::Conv_672 = Identity(%onnx::Conv_663)
%onnx::Conv_669 = Identity(%onnx::Conv_663)
%onnx::Conv_666 = Identity(%onnx::Conv_663)
%onnx::Conv_660 = Identity(%onnx::Conv_657)
%onnx::Conv_654 = Identity(%onnx::Conv_645)
%onnx::Conv_651 = Identity(%onnx::Conv_645)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_644, %onnx::Conv_645)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.3/Add_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%642 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %642
}
|
val_accuracy
| 0 | 58,668,672 | 1,952,764 |
{'zcp_synflow': 70.21926243565593, 'zcp_zen': 61.901065826416016, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.1158217340707779, 'zcp_flops': 58668672.0, 'zcp_grad_norm': 19.79058074951172, 'zcp_grasp': 0.027068138122558594, 'zcp_jacov': -16.052631690606415, 'zcp_l2_norm': 574.2167358398438, 'zcp_nwot': 209.08494130706674, 'zcp_params': 1952764.0, 'zcp_plain': -0.00306617165915668, 'zcp_snip': 34.292755126953125, 'lat_1080ti_1': 0.5353899192949907, 'lat_1080ti_32': 0.4656595936184217, 'lat_1080ti_64': 0.30291554574964025, 'lat_2080ti_1': 0.5320316418951816, 'lat_2080ti_32': 0.4384125515585064, 'lat_2080ti_64': 0.32519147046099156, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.32373975391047616, 'lat_fpga': 0.3518241819135357, 'lat_gold_6226': 0.2995378893166151, 'lat_gold_6240': 0.40101631062381987, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.29131499808906763, 'lat_raspi4': 0.366911695388212, 'lat_samsung_a50': 0.14736842105263157, 'lat_samsung_s7': 0.1968503937007874, 'lat_silver_4114': 0.4086420521557398, 'lat_silver_4210r': 0.37999223944767235, 'lat_titan_rtx_1': 0.4929700002183826, 'lat_titan_rtx_32': 0.42989434730115617, 'lat_titan_rtx_64': 0.3538279535908301, 'lat_titanx_1': 0.2596317181497743, 'lat_titanx_32': 0.3664301242635108, 'lat_titanx_64': 0.2933851539284589, 'lat_titanxp_1': 0.4657889021682886, 'lat_titanxp_32': 0.3862414466607338, 'lat_titanxp_64': 0.30817508633257407}
| |
FBNet_1349
|
FBNet
|
1349
|
1349
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_659[FLOAT, 16x3x3x3]
%onnx::Conv_660[FLOAT, 16]
%onnx::Conv_662[FLOAT, 16x8x1x1]
%onnx::Conv_665[FLOAT, 16x1x3x3]
%onnx::Conv_668[FLOAT, 16x8x1x1]
%onnx::Conv_671[FLOAT, 48x16x1x1]
%onnx::Conv_672[FLOAT, 48]
%onnx::Conv_674[FLOAT, 48x1x3x3]
%onnx::Conv_677[FLOAT, 24x48x1x1]
%onnx::Conv_678[FLOAT, 24]
%onnx::Conv_680[FLOAT, 144x24x1x1]
%onnx::Conv_681[FLOAT, 144]
%onnx::Conv_683[FLOAT, 144x1x5x5]
%onnx::Conv_686[FLOAT, 24x144x1x1]
%onnx::Conv_689[FLOAT, 144x24x1x1]
%onnx::Conv_692[FLOAT, 144x1x5x5]
%onnx::Conv_695[FLOAT, 24x144x1x1]
%onnx::Conv_698[FLOAT, 144x24x1x1]
%onnx::Conv_701[FLOAT, 144x1x5x5]
%onnx::Conv_704[FLOAT, 32x144x1x1]
%onnx::Conv_705[FLOAT, 32]
%onnx::Conv_707[FLOAT, 32x16x1x1]
%onnx::Conv_710[FLOAT, 32x1x5x5]
%onnx::Conv_713[FLOAT, 32x16x1x1]
%onnx::Conv_716[FLOAT, 32x32x1x1]
%onnx::Conv_719[FLOAT, 32x1x5x5]
%onnx::Conv_722[FLOAT, 32x32x1x1]
%onnx::Conv_725[FLOAT, 32x16x1x1]
%onnx::Conv_728[FLOAT, 32x1x3x3]
%onnx::Conv_731[FLOAT, 32x16x1x1]
%onnx::Conv_734[FLOAT, 96x32x1x1]
%onnx::Conv_735[FLOAT, 96]
%onnx::Conv_737[FLOAT, 96x1x3x3]
%onnx::Conv_740[FLOAT, 64x96x1x1]
%onnx::Conv_741[FLOAT, 64]
%onnx::Conv_743[FLOAT, 64x64x1x1]
%onnx::Conv_746[FLOAT, 64x1x5x5]
%onnx::Conv_749[FLOAT, 64x64x1x1]
%onnx::Conv_752[FLOAT, 192x64x1x1]
%onnx::Conv_753[FLOAT, 192]
%onnx::Conv_755[FLOAT, 192x1x5x5]
%onnx::Conv_758[FLOAT, 64x192x1x1]
%onnx::Conv_761[FLOAT, 64x32x1x1]
%onnx::Conv_764[FLOAT, 64x1x3x3]
%onnx::Conv_767[FLOAT, 64x32x1x1]
%onnx::Conv_770[FLOAT, 384x64x1x1]
%onnx::Conv_771[FLOAT, 384]
%onnx::Conv_773[FLOAT, 384x1x3x3]
%onnx::Conv_776[FLOAT, 112x384x1x1]
%onnx::Conv_777[FLOAT, 112]
%onnx::Conv_779[FLOAT, 672x112x1x1]
%onnx::Conv_780[FLOAT, 672]
%onnx::Conv_782[FLOAT, 672x1x3x3]
%onnx::Conv_785[FLOAT, 112x672x1x1]
%onnx::Conv_788[FLOAT, 112x112x1x1]
%onnx::Conv_791[FLOAT, 112x1x5x5]
%onnx::Conv_794[FLOAT, 112x112x1x1]
%onnx::Conv_797[FLOAT, 112x56x1x1]
%onnx::Conv_800[FLOAT, 112x1x5x5]
%onnx::Conv_803[FLOAT, 112x56x1x1]
%onnx::Conv_806[FLOAT, 672x112x1x1]
%onnx::Conv_809[FLOAT, 672x1x5x5]
%onnx::Conv_812[FLOAT, 184x672x1x1]
%onnx::Conv_813[FLOAT, 184]
%onnx::Conv_815[FLOAT, 552x184x1x1]
%onnx::Conv_816[FLOAT, 552]
%onnx::Conv_818[FLOAT, 552x1x3x3]
%onnx::Conv_821[FLOAT, 184x552x1x1]
%onnx::Conv_824[FLOAT, 1104x184x1x1]
%onnx::Conv_825[FLOAT, 1104]
%onnx::Conv_827[FLOAT, 1104x1x5x5]
%onnx::Conv_830[FLOAT, 184x1104x1x1]
%onnx::Conv_833[FLOAT, 184x184x1x1]
%onnx::Conv_836[FLOAT, 184x1x5x5]
%onnx::Conv_839[FLOAT, 352x184x1x1]
%onnx::Conv_840[FLOAT, 352]
%onnx::Conv_842[FLOAT, 1504x352x1x1]
%onnx::Conv_843[FLOAT, 1504]
) {
%onnx::Conv_837 = Identity(%onnx::Conv_813)
%onnx::Conv_834 = Identity(%onnx::Conv_813)
%onnx::Conv_831 = Identity(%onnx::Conv_813)
%onnx::Conv_828 = Identity(%onnx::Conv_825)
%onnx::Conv_822 = Identity(%onnx::Conv_813)
%onnx::Conv_819 = Identity(%onnx::Conv_816)
%onnx::Conv_810 = Identity(%onnx::Conv_780)
%onnx::Conv_807 = Identity(%onnx::Conv_780)
%onnx::Conv_804 = Identity(%onnx::Conv_777)
%onnx::Conv_801 = Identity(%onnx::Conv_777)
%onnx::Conv_798 = Identity(%onnx::Conv_777)
%onnx::Conv_795 = Identity(%onnx::Conv_777)
%onnx::Conv_792 = Identity(%onnx::Conv_777)
%onnx::Conv_789 = Identity(%onnx::Conv_777)
%onnx::Conv_786 = Identity(%onnx::Conv_777)
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_768 = Identity(%onnx::Conv_741)
%onnx::Conv_765 = Identity(%onnx::Conv_741)
%onnx::Conv_762 = Identity(%onnx::Conv_741)
%onnx::Conv_759 = Identity(%onnx::Conv_741)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_750 = Identity(%onnx::Conv_741)
%onnx::Conv_747 = Identity(%onnx::Conv_741)
%onnx::Conv_744 = Identity(%onnx::Conv_741)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_732 = Identity(%onnx::Conv_705)
%onnx::Conv_729 = Identity(%onnx::Conv_705)
%onnx::Conv_726 = Identity(%onnx::Conv_705)
%onnx::Conv_723 = Identity(%onnx::Conv_705)
%onnx::Conv_720 = Identity(%onnx::Conv_705)
%onnx::Conv_717 = Identity(%onnx::Conv_705)
%onnx::Conv_714 = Identity(%onnx::Conv_705)
%onnx::Conv_711 = Identity(%onnx::Conv_705)
%onnx::Conv_708 = Identity(%onnx::Conv_705)
%onnx::Conv_702 = Identity(%onnx::Conv_681)
%onnx::Conv_699 = Identity(%onnx::Conv_681)
%onnx::Conv_696 = Identity(%onnx::Conv_678)
%onnx::Conv_693 = Identity(%onnx::Conv_681)
%onnx::Conv_690 = Identity(%onnx::Conv_681)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_675 = Identity(%onnx::Conv_672)
%onnx::Conv_669 = Identity(%onnx::Conv_660)
%onnx::Conv_666 = Identity(%onnx::Conv_660)
%onnx::Conv_663 = Identity(%onnx::Conv_660)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_659, %onnx::Conv_660)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%657 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %657
}
|
val_accuracy
| 0 | 84,630,144 | 2,033,332 |
{'zcp_synflow': 76.35271203949287, 'zcp_zen': 68.30036163330078, 'zcp_epe_nas': 15.646582486687567, 'zcp_fisher': 0.10348396748304367, 'zcp_flops': 84630144.0, 'zcp_grad_norm': 25.224985122680664, 'zcp_grasp': -0.08485603332519531, 'zcp_jacov': -16.060623458177638, 'zcp_l2_norm': 633.5420532226562, 'zcp_nwot': 216.54017908250688, 'zcp_params': 2033332.0, 'zcp_plain': 0.00036034913500770926, 'zcp_snip': 43.83563232421875, 'lat_1080ti_1': 0.6589491383733014, 'lat_1080ti_32': 0.6817756361758515, 'lat_1080ti_64': 0.6688205911160983, 'lat_2080ti_1': 0.612729171646117, 'lat_2080ti_32': 0.7077576675819469, 'lat_2080ti_64': 0.7010957581177443, 'lat_essential_ph_1': 0.5094339622641509, 'lat_eyeriss': 0.6257770524269707, 'lat_fpga': 0.5856677782285594, 'lat_gold_6226': 0.42938496159315215, 'lat_gold_6240': 0.5053556860116479, 'lat_pixel2': 0.391304347826087, 'lat_pixel3': 0.6678244953816, 'lat_raspi4': 0.67390223726855, 'lat_samsung_a50': 0.25263157894736843, 'lat_samsung_s7': 0.2047244094488189, 'lat_silver_4114': 0.5933945362162977, 'lat_silver_4210r': 0.5223943612636018, 'lat_titan_rtx_1': 0.5878292774577466, 'lat_titan_rtx_32': 0.6794504374750737, 'lat_titan_rtx_64': 0.7070498221232624, 'lat_titanx_1': 0.31729424319805216, 'lat_titanx_32': 0.7167247391498736, 'lat_titanx_64': 0.7127830228376217, 'lat_titanxp_1': 0.5604845347793418, 'lat_titanxp_32': 0.7197713652028652, 'lat_titanxp_64': 0.704915005918368}
| |
FBNet_2875
|
FBNet
|
2875
|
2875
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_687[FLOAT, 16x3x3x3]
%onnx::Conv_688[FLOAT, 16]
%onnx::Conv_690[FLOAT, 16x16x1x1]
%onnx::Conv_693[FLOAT, 16x1x5x5]
%onnx::Conv_696[FLOAT, 16x16x1x1]
%onnx::Conv_699[FLOAT, 16x16x1x1]
%onnx::Conv_702[FLOAT, 16x1x3x3]
%onnx::Conv_705[FLOAT, 24x16x1x1]
%onnx::Conv_706[FLOAT, 24]
%onnx::Conv_708[FLOAT, 24x12x1x1]
%onnx::Conv_711[FLOAT, 24x1x3x3]
%onnx::Conv_714[FLOAT, 24x12x1x1]
%onnx::Conv_717[FLOAT, 24x12x1x1]
%onnx::Conv_720[FLOAT, 24x1x5x5]
%onnx::Conv_723[FLOAT, 24x12x1x1]
%onnx::Conv_726[FLOAT, 24x24x1x1]
%onnx::Conv_729[FLOAT, 24x1x3x3]
%onnx::Conv_732[FLOAT, 24x24x1x1]
%onnx::Conv_735[FLOAT, 72x24x1x1]
%onnx::Conv_736[FLOAT, 72]
%onnx::Conv_738[FLOAT, 72x1x3x3]
%onnx::Conv_741[FLOAT, 32x72x1x1]
%onnx::Conv_742[FLOAT, 32]
%onnx::Conv_744[FLOAT, 96x32x1x1]
%onnx::Conv_745[FLOAT, 96]
%onnx::Conv_747[FLOAT, 96x1x5x5]
%onnx::Conv_750[FLOAT, 32x96x1x1]
%onnx::Conv_753[FLOAT, 32x16x1x1]
%onnx::Conv_756[FLOAT, 32x1x3x3]
%onnx::Conv_759[FLOAT, 32x16x1x1]
%onnx::Conv_762[FLOAT, 192x32x1x1]
%onnx::Conv_763[FLOAT, 192]
%onnx::Conv_765[FLOAT, 192x1x5x5]
%onnx::Conv_768[FLOAT, 32x192x1x1]
%onnx::Conv_771[FLOAT, 32x16x1x1]
%onnx::Conv_774[FLOAT, 32x1x3x3]
%onnx::Conv_777[FLOAT, 64x16x1x1]
%onnx::Conv_778[FLOAT, 64]
%onnx::Conv_780[FLOAT, 192x64x1x1]
%onnx::Conv_783[FLOAT, 192x1x3x3]
%onnx::Conv_786[FLOAT, 64x192x1x1]
%onnx::Conv_789[FLOAT, 192x64x1x1]
%onnx::Conv_792[FLOAT, 192x1x3x3]
%onnx::Conv_795[FLOAT, 64x192x1x1]
%onnx::Conv_798[FLOAT, 64x32x1x1]
%onnx::Conv_801[FLOAT, 64x1x5x5]
%onnx::Conv_804[FLOAT, 64x32x1x1]
%onnx::Conv_807[FLOAT, 112x64x1x1]
%onnx::Conv_808[FLOAT, 112]
%onnx::Conv_810[FLOAT, 112x56x1x1]
%onnx::Conv_813[FLOAT, 112x1x3x3]
%onnx::Conv_816[FLOAT, 112x56x1x1]
%onnx::Conv_819[FLOAT, 672x112x1x1]
%onnx::Conv_820[FLOAT, 672]
%onnx::Conv_822[FLOAT, 672x1x5x5]
%onnx::Conv_825[FLOAT, 112x672x1x1]
%onnx::Conv_828[FLOAT, 672x112x1x1]
%onnx::Conv_831[FLOAT, 672x1x5x5]
%onnx::Conv_834[FLOAT, 184x672x1x1]
%onnx::Conv_835[FLOAT, 184]
%onnx::Conv_837[FLOAT, 552x184x1x1]
%onnx::Conv_838[FLOAT, 552]
%onnx::Conv_840[FLOAT, 552x1x5x5]
%onnx::Conv_843[FLOAT, 184x552x1x1]
%onnx::Conv_846[FLOAT, 184x184x1x1]
%onnx::Conv_849[FLOAT, 184x1x3x3]
%onnx::Conv_852[FLOAT, 184x184x1x1]
%onnx::Conv_855[FLOAT, 1104x184x1x1]
%onnx::Conv_856[FLOAT, 1104]
%onnx::Conv_858[FLOAT, 1104x1x5x5]
%onnx::Conv_861[FLOAT, 184x1104x1x1]
%onnx::Conv_864[FLOAT, 184x184x1x1]
%onnx::Conv_867[FLOAT, 184x1x3x3]
%onnx::Conv_870[FLOAT, 352x184x1x1]
%onnx::Conv_871[FLOAT, 352]
%onnx::Conv_873[FLOAT, 1504x352x1x1]
%onnx::Conv_874[FLOAT, 1504]
) {
%onnx::Conv_868 = Identity(%onnx::Conv_835)
%onnx::Conv_865 = Identity(%onnx::Conv_835)
%onnx::Conv_862 = Identity(%onnx::Conv_835)
%onnx::Conv_859 = Identity(%onnx::Conv_856)
%onnx::Conv_853 = Identity(%onnx::Conv_835)
%onnx::Conv_850 = Identity(%onnx::Conv_835)
%onnx::Conv_847 = Identity(%onnx::Conv_835)
%onnx::Conv_844 = Identity(%onnx::Conv_835)
%onnx::Conv_841 = Identity(%onnx::Conv_838)
%onnx::Conv_832 = Identity(%onnx::Conv_820)
%onnx::Conv_829 = Identity(%onnx::Conv_820)
%onnx::Conv_826 = Identity(%onnx::Conv_808)
%onnx::Conv_823 = Identity(%onnx::Conv_820)
%onnx::Conv_817 = Identity(%onnx::Conv_808)
%onnx::Conv_814 = Identity(%onnx::Conv_808)
%onnx::Conv_811 = Identity(%onnx::Conv_808)
%onnx::Conv_805 = Identity(%onnx::Conv_778)
%onnx::Conv_802 = Identity(%onnx::Conv_778)
%onnx::Conv_799 = Identity(%onnx::Conv_778)
%onnx::Conv_796 = Identity(%onnx::Conv_778)
%onnx::Conv_793 = Identity(%onnx::Conv_763)
%onnx::Conv_790 = Identity(%onnx::Conv_763)
%onnx::Conv_787 = Identity(%onnx::Conv_778)
%onnx::Conv_784 = Identity(%onnx::Conv_763)
%onnx::Conv_781 = Identity(%onnx::Conv_763)
%onnx::Conv_775 = Identity(%onnx::Conv_742)
%onnx::Conv_772 = Identity(%onnx::Conv_742)
%onnx::Conv_769 = Identity(%onnx::Conv_742)
%onnx::Conv_766 = Identity(%onnx::Conv_763)
%onnx::Conv_760 = Identity(%onnx::Conv_742)
%onnx::Conv_757 = Identity(%onnx::Conv_742)
%onnx::Conv_754 = Identity(%onnx::Conv_742)
%onnx::Conv_751 = Identity(%onnx::Conv_742)
%onnx::Conv_748 = Identity(%onnx::Conv_745)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_733 = Identity(%onnx::Conv_706)
%onnx::Conv_730 = Identity(%onnx::Conv_706)
%onnx::Conv_727 = Identity(%onnx::Conv_706)
%onnx::Conv_724 = Identity(%onnx::Conv_706)
%onnx::Conv_721 = Identity(%onnx::Conv_706)
%onnx::Conv_718 = Identity(%onnx::Conv_706)
%onnx::Conv_715 = Identity(%onnx::Conv_706)
%onnx::Conv_712 = Identity(%onnx::Conv_706)
%onnx::Conv_709 = Identity(%onnx::Conv_706)
%onnx::Conv_703 = Identity(%onnx::Conv_688)
%onnx::Conv_700 = Identity(%onnx::Conv_688)
%onnx::Conv_697 = Identity(%onnx::Conv_688)
%onnx::Conv_694 = Identity(%onnx::Conv_688)
%onnx::Conv_691 = Identity(%onnx::Conv_688)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_687, %onnx::Conv_688)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%685 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %685
}
|
val_accuracy
| 0 | 62,211,712 | 2,026,044 |
{'zcp_synflow': 75.88352645167318, 'zcp_zen': 66.07376861572266, 'zcp_epe_nas': 7.8965389976761795, 'zcp_fisher': 0.1302117258310318, 'zcp_flops': 62211712.0, 'zcp_grad_norm': 22.542673110961914, 'zcp_grasp': -0.040058135986328125, 'zcp_jacov': -16.054919143564607, 'zcp_l2_norm': 612.8863525390625, 'zcp_nwot': 208.22264201014323, 'zcp_params': 2026044.0, 'zcp_plain': -0.008528807200491428, 'zcp_snip': 36.8766975402832, 'lat_1080ti_1': 0.6006834037214311, 'lat_1080ti_32': 0.5415557151436253, 'lat_1080ti_64': 0.3360685752094373, 'lat_2080ti_1': 0.6772123961300697, 'lat_2080ti_32': 0.5328521293524957, 'lat_2080ti_64': 0.33443302084217413, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.36493499349018016, 'lat_fpga': 0.3581175353011606, 'lat_gold_6226': 0.3766700756494929, 'lat_gold_6240': 0.5653181134579578, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.3717401375829788, 'lat_raspi4': 0.3950077206401519, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.1732283464566929, 'lat_silver_4114': 0.5780245770652087, 'lat_silver_4210r': 0.6064452796281066, 'lat_titan_rtx_1': 0.6502786247344716, 'lat_titan_rtx_32': 0.5235662029762154, 'lat_titan_rtx_64': 0.39209714830081127, 'lat_titanx_1': 0.34520664462675593, 'lat_titanx_32': 0.4125229502003237, 'lat_titanx_64': 0.36833736190691047, 'lat_titanxp_1': 0.6169647991227545, 'lat_titanxp_32': 0.4840544547184334, 'lat_titanxp_64': 0.3318967344567913}
| |
FBNet_3506
|
FBNet
|
3506
|
3506
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_687[FLOAT, 16x3x3x3]
%onnx::Conv_688[FLOAT, 16]
%onnx::Conv_690[FLOAT, 16x8x1x1]
%onnx::Conv_693[FLOAT, 16x1x5x5]
%onnx::Conv_696[FLOAT, 16x8x1x1]
%onnx::Conv_699[FLOAT, 16x8x1x1]
%onnx::Conv_702[FLOAT, 16x1x3x3]
%onnx::Conv_705[FLOAT, 24x8x1x1]
%onnx::Conv_706[FLOAT, 24]
%onnx::Conv_708[FLOAT, 144x24x1x1]
%onnx::Conv_709[FLOAT, 144]
%onnx::Conv_711[FLOAT, 144x1x3x3]
%onnx::Conv_714[FLOAT, 24x144x1x1]
%onnx::Conv_717[FLOAT, 24x12x1x1]
%onnx::Conv_720[FLOAT, 24x1x3x3]
%onnx::Conv_723[FLOAT, 24x12x1x1]
%onnx::Conv_726[FLOAT, 24x24x1x1]
%onnx::Conv_729[FLOAT, 24x1x5x5]
%onnx::Conv_732[FLOAT, 32x24x1x1]
%onnx::Conv_733[FLOAT, 32]
%onnx::Conv_735[FLOAT, 32x16x1x1]
%onnx::Conv_738[FLOAT, 32x1x5x5]
%onnx::Conv_741[FLOAT, 32x16x1x1]
%onnx::Conv_744[FLOAT, 32x16x1x1]
%onnx::Conv_747[FLOAT, 32x1x3x3]
%onnx::Conv_750[FLOAT, 32x16x1x1]
%onnx::Conv_753[FLOAT, 192x32x1x1]
%onnx::Conv_754[FLOAT, 192]
%onnx::Conv_756[FLOAT, 192x1x5x5]
%onnx::Conv_759[FLOAT, 32x192x1x1]
%onnx::Conv_762[FLOAT, 32x32x1x1]
%onnx::Conv_765[FLOAT, 32x1x3x3]
%onnx::Conv_768[FLOAT, 64x32x1x1]
%onnx::Conv_769[FLOAT, 64]
%onnx::Conv_771[FLOAT, 64x64x1x1]
%onnx::Conv_774[FLOAT, 64x1x3x3]
%onnx::Conv_777[FLOAT, 64x64x1x1]
%onnx::Conv_780[FLOAT, 384x64x1x1]
%onnx::Conv_781[FLOAT, 384]
%onnx::Conv_783[FLOAT, 384x1x5x5]
%onnx::Conv_786[FLOAT, 64x384x1x1]
%onnx::Conv_789[FLOAT, 64x32x1x1]
%onnx::Conv_792[FLOAT, 64x1x5x5]
%onnx::Conv_795[FLOAT, 112x32x1x1]
%onnx::Conv_796[FLOAT, 112]
%onnx::Conv_798[FLOAT, 112x56x1x1]
%onnx::Conv_801[FLOAT, 112x1x3x3]
%onnx::Conv_804[FLOAT, 112x56x1x1]
%onnx::Conv_807[FLOAT, 672x112x1x1]
%onnx::Conv_808[FLOAT, 672]
%onnx::Conv_810[FLOAT, 672x1x5x5]
%onnx::Conv_813[FLOAT, 112x672x1x1]
%onnx::Conv_816[FLOAT, 112x56x1x1]
%onnx::Conv_819[FLOAT, 112x1x3x3]
%onnx::Conv_822[FLOAT, 112x56x1x1]
%onnx::Conv_825[FLOAT, 672x112x1x1]
%onnx::Conv_828[FLOAT, 672x1x3x3]
%onnx::Conv_831[FLOAT, 184x672x1x1]
%onnx::Conv_832[FLOAT, 184]
%onnx::Conv_834[FLOAT, 184x184x1x1]
%onnx::Conv_837[FLOAT, 184x1x3x3]
%onnx::Conv_840[FLOAT, 184x184x1x1]
%onnx::Conv_843[FLOAT, 184x184x1x1]
%onnx::Conv_846[FLOAT, 184x1x3x3]
%onnx::Conv_849[FLOAT, 184x184x1x1]
%onnx::Conv_852[FLOAT, 552x184x1x1]
%onnx::Conv_853[FLOAT, 552]
%onnx::Conv_855[FLOAT, 552x1x3x3]
%onnx::Conv_858[FLOAT, 352x552x1x1]
%onnx::Conv_859[FLOAT, 352]
%onnx::Conv_861[FLOAT, 1504x352x1x1]
%onnx::Conv_862[FLOAT, 1504]
) {
%onnx::Conv_856 = Identity(%onnx::Conv_853)
%onnx::Conv_850 = Identity(%onnx::Conv_832)
%onnx::Conv_847 = Identity(%onnx::Conv_832)
%onnx::Conv_844 = Identity(%onnx::Conv_832)
%onnx::Conv_841 = Identity(%onnx::Conv_832)
%onnx::Conv_838 = Identity(%onnx::Conv_832)
%onnx::Conv_835 = Identity(%onnx::Conv_832)
%onnx::Conv_829 = Identity(%onnx::Conv_808)
%onnx::Conv_826 = Identity(%onnx::Conv_808)
%onnx::Conv_823 = Identity(%onnx::Conv_796)
%onnx::Conv_820 = Identity(%onnx::Conv_796)
%onnx::Conv_817 = Identity(%onnx::Conv_796)
%onnx::Conv_814 = Identity(%onnx::Conv_796)
%onnx::Conv_811 = Identity(%onnx::Conv_808)
%onnx::Conv_805 = Identity(%onnx::Conv_796)
%onnx::Conv_802 = Identity(%onnx::Conv_796)
%onnx::Conv_799 = Identity(%onnx::Conv_796)
%onnx::Conv_793 = Identity(%onnx::Conv_769)
%onnx::Conv_790 = Identity(%onnx::Conv_769)
%onnx::Conv_787 = Identity(%onnx::Conv_769)
%onnx::Conv_784 = Identity(%onnx::Conv_781)
%onnx::Conv_778 = Identity(%onnx::Conv_769)
%onnx::Conv_775 = Identity(%onnx::Conv_769)
%onnx::Conv_772 = Identity(%onnx::Conv_769)
%onnx::Conv_766 = Identity(%onnx::Conv_733)
%onnx::Conv_763 = Identity(%onnx::Conv_733)
%onnx::Conv_760 = Identity(%onnx::Conv_733)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_751 = Identity(%onnx::Conv_733)
%onnx::Conv_748 = Identity(%onnx::Conv_733)
%onnx::Conv_745 = Identity(%onnx::Conv_733)
%onnx::Conv_742 = Identity(%onnx::Conv_733)
%onnx::Conv_739 = Identity(%onnx::Conv_733)
%onnx::Conv_736 = Identity(%onnx::Conv_733)
%onnx::Conv_730 = Identity(%onnx::Conv_706)
%onnx::Conv_727 = Identity(%onnx::Conv_706)
%onnx::Conv_724 = Identity(%onnx::Conv_706)
%onnx::Conv_721 = Identity(%onnx::Conv_706)
%onnx::Conv_718 = Identity(%onnx::Conv_706)
%onnx::Conv_715 = Identity(%onnx::Conv_706)
%onnx::Conv_712 = Identity(%onnx::Conv_709)
%onnx::Conv_703 = Identity(%onnx::Conv_688)
%onnx::Conv_700 = Identity(%onnx::Conv_688)
%onnx::Conv_697 = Identity(%onnx::Conv_688)
%onnx::Conv_694 = Identity(%onnx::Conv_688)
%onnx::Conv_691 = Identity(%onnx::Conv_688)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_687, %onnx::Conv_688)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%685 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %685
}
|
val_accuracy
| 0 | 59,694,464 | 1,650,316 |
{'zcp_synflow': 67.24190940591164, 'zcp_zen': 59.92041778564453, 'zcp_epe_nas': 8.078465470664728, 'zcp_fisher': 0.08455692231655121, 'zcp_flops': 59694464.0, 'zcp_grad_norm': 20.02983283996582, 'zcp_grasp': -0.34522247314453125, 'zcp_jacov': -16.057984161899213, 'zcp_l2_norm': 538.2520751953125, 'zcp_nwot': 209.5721399151106, 'zcp_params': 1650316.0, 'zcp_plain': 0.0028500284533947706, 'zcp_snip': 31.22237777709961, 'lat_1080ti_1': 0.515202401389786, 'lat_1080ti_32': 0.483810358663802, 'lat_1080ti_64': 0.3539170425478081, 'lat_2080ti_1': 0.6063409244727164, 'lat_2080ti_32': 0.5261353373498315, 'lat_2080ti_64': 0.4096276393122684, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.3069700915042267, 'lat_fpga': 0.337060528997912, 'lat_gold_6226': 0.24943109330149732, 'lat_gold_6240': 0.3867239903862073, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.34821961427196535, 'lat_raspi4': 0.40076116005156837, 'lat_samsung_a50': 0.1368421052631579, 'lat_samsung_s7': 0.12598425196850394, 'lat_silver_4114': 0.42460276881747916, 'lat_silver_4210r': 0.4470812312686828, 'lat_titan_rtx_1': 0.5506810342779835, 'lat_titan_rtx_32': 0.49981280546552187, 'lat_titan_rtx_64': 0.4351654343429162, 'lat_titanx_1': 0.29493835947581704, 'lat_titanx_32': 0.44932646011302646, 'lat_titanx_64': 0.38657530888577496, 'lat_titanxp_1': 0.5107753246578858, 'lat_titanxp_32': 0.4807033067126037, 'lat_titanxp_64': 0.38553377560901236}
| |
FBNet_2783
|
FBNet
|
2783
|
2783
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_658[FLOAT, 16x3x3x3]
%onnx::Conv_659[FLOAT, 16]
%onnx::Conv_661[FLOAT, 48x16x1x1]
%onnx::Conv_662[FLOAT, 48]
%onnx::Conv_664[FLOAT, 48x1x5x5]
%onnx::Conv_667[FLOAT, 16x48x1x1]
%onnx::Conv_670[FLOAT, 48x16x1x1]
%onnx::Conv_673[FLOAT, 48x1x5x5]
%onnx::Conv_676[FLOAT, 24x48x1x1]
%onnx::Conv_677[FLOAT, 24]
%onnx::Conv_679[FLOAT, 72x24x1x1]
%onnx::Conv_680[FLOAT, 72]
%onnx::Conv_682[FLOAT, 72x1x5x5]
%onnx::Conv_685[FLOAT, 24x72x1x1]
%onnx::Conv_688[FLOAT, 72x24x1x1]
%onnx::Conv_691[FLOAT, 72x1x5x5]
%onnx::Conv_694[FLOAT, 24x72x1x1]
%onnx::Conv_697[FLOAT, 72x24x1x1]
%onnx::Conv_700[FLOAT, 72x1x3x3]
%onnx::Conv_703[FLOAT, 32x72x1x1]
%onnx::Conv_704[FLOAT, 32]
%onnx::Conv_706[FLOAT, 96x32x1x1]
%onnx::Conv_707[FLOAT, 96]
%onnx::Conv_709[FLOAT, 96x1x5x5]
%onnx::Conv_712[FLOAT, 32x96x1x1]
%onnx::Conv_715[FLOAT, 192x32x1x1]
%onnx::Conv_716[FLOAT, 192]
%onnx::Conv_718[FLOAT, 192x1x3x3]
%onnx::Conv_721[FLOAT, 32x192x1x1]
%onnx::Conv_724[FLOAT, 96x32x1x1]
%onnx::Conv_727[FLOAT, 96x1x3x3]
%onnx::Conv_730[FLOAT, 32x96x1x1]
%onnx::Conv_733[FLOAT, 32x32x1x1]
%onnx::Conv_736[FLOAT, 32x1x3x3]
%onnx::Conv_739[FLOAT, 64x32x1x1]
%onnx::Conv_740[FLOAT, 64]
%onnx::Conv_742[FLOAT, 384x64x1x1]
%onnx::Conv_743[FLOAT, 384]
%onnx::Conv_745[FLOAT, 384x1x5x5]
%onnx::Conv_748[FLOAT, 64x384x1x1]
%onnx::Conv_751[FLOAT, 384x64x1x1]
%onnx::Conv_754[FLOAT, 384x1x3x3]
%onnx::Conv_757[FLOAT, 64x384x1x1]
%onnx::Conv_760[FLOAT, 64x32x1x1]
%onnx::Conv_763[FLOAT, 64x1x5x5]
%onnx::Conv_766[FLOAT, 112x32x1x1]
%onnx::Conv_767[FLOAT, 112]
%onnx::Conv_769[FLOAT, 112x56x1x1]
%onnx::Conv_772[FLOAT, 112x1x5x5]
%onnx::Conv_775[FLOAT, 112x56x1x1]
%onnx::Conv_778[FLOAT, 112x56x1x1]
%onnx::Conv_781[FLOAT, 112x1x3x3]
%onnx::Conv_784[FLOAT, 112x56x1x1]
%onnx::Conv_787[FLOAT, 112x56x1x1]
%onnx::Conv_790[FLOAT, 112x1x5x5]
%onnx::Conv_793[FLOAT, 112x56x1x1]
%onnx::Conv_796[FLOAT, 672x112x1x1]
%onnx::Conv_797[FLOAT, 672]
%onnx::Conv_799[FLOAT, 672x1x5x5]
%onnx::Conv_802[FLOAT, 184x672x1x1]
%onnx::Conv_803[FLOAT, 184]
%onnx::Conv_805[FLOAT, 184x184x1x1]
%onnx::Conv_808[FLOAT, 184x1x5x5]
%onnx::Conv_811[FLOAT, 184x184x1x1]
%onnx::Conv_814[FLOAT, 552x184x1x1]
%onnx::Conv_815[FLOAT, 552]
%onnx::Conv_817[FLOAT, 552x1x5x5]
%onnx::Conv_820[FLOAT, 184x552x1x1]
%onnx::Conv_823[FLOAT, 184x92x1x1]
%onnx::Conv_826[FLOAT, 184x1x5x5]
%onnx::Conv_829[FLOAT, 184x92x1x1]
%onnx::Conv_832[FLOAT, 184x184x1x1]
%onnx::Conv_835[FLOAT, 184x1x3x3]
%onnx::Conv_838[FLOAT, 352x184x1x1]
%onnx::Conv_839[FLOAT, 352]
%onnx::Conv_841[FLOAT, 1504x352x1x1]
%onnx::Conv_842[FLOAT, 1504]
) {
%onnx::Conv_836 = Identity(%onnx::Conv_803)
%onnx::Conv_833 = Identity(%onnx::Conv_803)
%onnx::Conv_830 = Identity(%onnx::Conv_803)
%onnx::Conv_827 = Identity(%onnx::Conv_803)
%onnx::Conv_824 = Identity(%onnx::Conv_803)
%onnx::Conv_821 = Identity(%onnx::Conv_803)
%onnx::Conv_818 = Identity(%onnx::Conv_815)
%onnx::Conv_812 = Identity(%onnx::Conv_803)
%onnx::Conv_809 = Identity(%onnx::Conv_803)
%onnx::Conv_806 = Identity(%onnx::Conv_803)
%onnx::Conv_800 = Identity(%onnx::Conv_797)
%onnx::Conv_794 = Identity(%onnx::Conv_767)
%onnx::Conv_791 = Identity(%onnx::Conv_767)
%onnx::Conv_788 = Identity(%onnx::Conv_767)
%onnx::Conv_785 = Identity(%onnx::Conv_767)
%onnx::Conv_782 = Identity(%onnx::Conv_767)
%onnx::Conv_779 = Identity(%onnx::Conv_767)
%onnx::Conv_776 = Identity(%onnx::Conv_767)
%onnx::Conv_773 = Identity(%onnx::Conv_767)
%onnx::Conv_770 = Identity(%onnx::Conv_767)
%onnx::Conv_764 = Identity(%onnx::Conv_740)
%onnx::Conv_761 = Identity(%onnx::Conv_740)
%onnx::Conv_758 = Identity(%onnx::Conv_740)
%onnx::Conv_755 = Identity(%onnx::Conv_743)
%onnx::Conv_752 = Identity(%onnx::Conv_743)
%onnx::Conv_749 = Identity(%onnx::Conv_740)
%onnx::Conv_746 = Identity(%onnx::Conv_743)
%onnx::Conv_737 = Identity(%onnx::Conv_704)
%onnx::Conv_734 = Identity(%onnx::Conv_704)
%onnx::Conv_731 = Identity(%onnx::Conv_704)
%onnx::Conv_728 = Identity(%onnx::Conv_707)
%onnx::Conv_725 = Identity(%onnx::Conv_707)
%onnx::Conv_722 = Identity(%onnx::Conv_704)
%onnx::Conv_719 = Identity(%onnx::Conv_716)
%onnx::Conv_713 = Identity(%onnx::Conv_704)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_701 = Identity(%onnx::Conv_680)
%onnx::Conv_698 = Identity(%onnx::Conv_680)
%onnx::Conv_695 = Identity(%onnx::Conv_677)
%onnx::Conv_692 = Identity(%onnx::Conv_680)
%onnx::Conv_689 = Identity(%onnx::Conv_680)
%onnx::Conv_686 = Identity(%onnx::Conv_677)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_674 = Identity(%onnx::Conv_662)
%onnx::Conv_671 = Identity(%onnx::Conv_662)
%onnx::Conv_668 = Identity(%onnx::Conv_659)
%onnx::Conv_665 = Identity(%onnx::Conv_662)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_658, %onnx::Conv_659)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%656 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %656
}
|
val_accuracy
| 0 | 63,697,280 | 1,562,460 |
{'zcp_synflow': 76.69246018991828, 'zcp_zen': 68.1550064086914, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.13812053203582764, 'zcp_flops': 63697280.0, 'zcp_grad_norm': 25.883134841918945, 'zcp_grasp': -0.11285400390625, 'zcp_jacov': -16.063679136743893, 'zcp_l2_norm': 598.9216918945312, 'zcp_nwot': 212.42724702400122, 'zcp_params': 1562460.0, 'zcp_plain': -0.0017299385508522391, 'zcp_snip': 46.99128723144531, 'lat_1080ti_1': 0.5290396834888424, 'lat_1080ti_32': 0.5717885872601836, 'lat_1080ti_64': 0.4811882222891349, 'lat_2080ti_1': 0.6129510631234723, 'lat_2080ti_32': 0.5258778465006481, 'lat_2080ti_64': 0.4686106188399606, 'lat_essential_ph_1': 0.3018867924528302, 'lat_eyeriss': 0.4307850292483451, 'lat_fpga': 0.3330232834284922, 'lat_gold_6226': 0.271863921699917, 'lat_gold_6240': 0.3776176516007667, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.42476447901288716, 'lat_raspi4': 0.36986536083672955, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.14960629921259844, 'lat_silver_4114': 0.4858785730135448, 'lat_silver_4210r': 0.3703312218665994, 'lat_titan_rtx_1': 0.5829054439098135, 'lat_titan_rtx_32': 0.5080758689874215, 'lat_titan_rtx_64': 0.48333930552127574, 'lat_titanx_1': 0.3101979012935683, 'lat_titanx_32': 0.4910800407869162, 'lat_titanx_64': 0.4981903459844006, 'lat_titanxp_1': 0.5375665461373496, 'lat_titanxp_32': 0.5143282447127823, 'lat_titanxp_64': 0.4862588414425173}
| |
FBNet_33
|
FBNet
|
33
|
33
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_569[FLOAT, 16x3x3x3]
%onnx::Conv_570[FLOAT, 16]
%onnx::Conv_572[FLOAT, 16x8x1x1]
%onnx::Conv_575[FLOAT, 16x1x3x3]
%onnx::Conv_578[FLOAT, 24x8x1x1]
%onnx::Conv_579[FLOAT, 24]
%onnx::Conv_581[FLOAT, 24x24x1x1]
%onnx::Conv_584[FLOAT, 24x1x5x5]
%onnx::Conv_587[FLOAT, 24x24x1x1]
%onnx::Conv_590[FLOAT, 24x24x1x1]
%onnx::Conv_593[FLOAT, 24x1x5x5]
%onnx::Conv_596[FLOAT, 24x24x1x1]
%onnx::Conv_599[FLOAT, 24x24x1x1]
%onnx::Conv_602[FLOAT, 24x1x3x3]
%onnx::Conv_605[FLOAT, 32x24x1x1]
%onnx::Conv_606[FLOAT, 32]
%onnx::Conv_608[FLOAT, 192x32x1x1]
%onnx::Conv_609[FLOAT, 192]
%onnx::Conv_611[FLOAT, 192x1x3x3]
%onnx::Conv_614[FLOAT, 32x192x1x1]
%onnx::Conv_617[FLOAT, 32x16x1x1]
%onnx::Conv_620[FLOAT, 32x1x5x5]
%onnx::Conv_623[FLOAT, 32x16x1x1]
%onnx::Conv_626[FLOAT, 64x32x1x1]
%onnx::Conv_627[FLOAT, 64]
%onnx::Conv_629[FLOAT, 384x64x1x1]
%onnx::Conv_630[FLOAT, 384]
%onnx::Conv_632[FLOAT, 384x1x3x3]
%onnx::Conv_635[FLOAT, 64x384x1x1]
%onnx::Conv_638[FLOAT, 192x64x1x1]
%onnx::Conv_641[FLOAT, 192x1x5x5]
%onnx::Conv_644[FLOAT, 64x192x1x1]
%onnx::Conv_647[FLOAT, 384x64x1x1]
%onnx::Conv_650[FLOAT, 384x1x3x3]
%onnx::Conv_653[FLOAT, 64x384x1x1]
%onnx::Conv_656[FLOAT, 192x64x1x1]
%onnx::Conv_659[FLOAT, 192x1x3x3]
%onnx::Conv_662[FLOAT, 112x192x1x1]
%onnx::Conv_663[FLOAT, 112]
%onnx::Conv_665[FLOAT, 336x112x1x1]
%onnx::Conv_666[FLOAT, 336]
%onnx::Conv_668[FLOAT, 336x1x3x3]
%onnx::Conv_671[FLOAT, 112x336x1x1]
%onnx::Conv_674[FLOAT, 112x56x1x1]
%onnx::Conv_677[FLOAT, 112x1x3x3]
%onnx::Conv_680[FLOAT, 112x56x1x1]
%onnx::Conv_683[FLOAT, 112x112x1x1]
%onnx::Conv_686[FLOAT, 112x1x5x5]
%onnx::Conv_689[FLOAT, 112x112x1x1]
%onnx::Conv_692[FLOAT, 112x56x1x1]
%onnx::Conv_695[FLOAT, 112x1x5x5]
%onnx::Conv_698[FLOAT, 184x56x1x1]
%onnx::Conv_699[FLOAT, 184]
%onnx::Conv_701[FLOAT, 552x184x1x1]
%onnx::Conv_702[FLOAT, 552]
%onnx::Conv_704[FLOAT, 552x1x5x5]
%onnx::Conv_707[FLOAT, 184x552x1x1]
%onnx::Conv_710[FLOAT, 184x184x1x1]
%onnx::Conv_713[FLOAT, 184x1x3x3]
%onnx::Conv_716[FLOAT, 184x184x1x1]
%onnx::Conv_719[FLOAT, 184x184x1x1]
%onnx::Conv_722[FLOAT, 184x1x3x3]
%onnx::Conv_725[FLOAT, 352x184x1x1]
%onnx::Conv_726[FLOAT, 352]
%onnx::Conv_728[FLOAT, 1504x352x1x1]
%onnx::Conv_729[FLOAT, 1504]
) {
%onnx::Conv_723 = Identity(%onnx::Conv_699)
%onnx::Conv_720 = Identity(%onnx::Conv_699)
%onnx::Conv_717 = Identity(%onnx::Conv_699)
%onnx::Conv_714 = Identity(%onnx::Conv_699)
%onnx::Conv_711 = Identity(%onnx::Conv_699)
%onnx::Conv_708 = Identity(%onnx::Conv_699)
%onnx::Conv_705 = Identity(%onnx::Conv_702)
%onnx::Conv_696 = Identity(%onnx::Conv_663)
%onnx::Conv_693 = Identity(%onnx::Conv_663)
%onnx::Conv_690 = Identity(%onnx::Conv_663)
%onnx::Conv_687 = Identity(%onnx::Conv_663)
%onnx::Conv_684 = Identity(%onnx::Conv_663)
%onnx::Conv_681 = Identity(%onnx::Conv_663)
%onnx::Conv_678 = Identity(%onnx::Conv_663)
%onnx::Conv_675 = Identity(%onnx::Conv_663)
%onnx::Conv_672 = Identity(%onnx::Conv_663)
%onnx::Conv_669 = Identity(%onnx::Conv_666)
%onnx::Conv_660 = Identity(%onnx::Conv_609)
%onnx::Conv_657 = Identity(%onnx::Conv_609)
%onnx::Conv_654 = Identity(%onnx::Conv_627)
%onnx::Conv_651 = Identity(%onnx::Conv_630)
%onnx::Conv_648 = Identity(%onnx::Conv_630)
%onnx::Conv_645 = Identity(%onnx::Conv_627)
%onnx::Conv_642 = Identity(%onnx::Conv_609)
%onnx::Conv_639 = Identity(%onnx::Conv_609)
%onnx::Conv_636 = Identity(%onnx::Conv_627)
%onnx::Conv_633 = Identity(%onnx::Conv_630)
%onnx::Conv_624 = Identity(%onnx::Conv_606)
%onnx::Conv_621 = Identity(%onnx::Conv_606)
%onnx::Conv_618 = Identity(%onnx::Conv_606)
%onnx::Conv_615 = Identity(%onnx::Conv_606)
%onnx::Conv_612 = Identity(%onnx::Conv_609)
%onnx::Conv_603 = Identity(%onnx::Conv_579)
%onnx::Conv_600 = Identity(%onnx::Conv_579)
%onnx::Conv_597 = Identity(%onnx::Conv_579)
%onnx::Conv_594 = Identity(%onnx::Conv_579)
%onnx::Conv_591 = Identity(%onnx::Conv_579)
%onnx::Conv_588 = Identity(%onnx::Conv_579)
%onnx::Conv_585 = Identity(%onnx::Conv_579)
%onnx::Conv_582 = Identity(%onnx::Conv_579)
%onnx::Conv_576 = Identity(%onnx::Conv_570)
%onnx::Conv_573 = Identity(%onnx::Conv_570)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_569, %onnx::Conv_570)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_572, %onnx::Conv_573)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_575, %onnx::Conv_576)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_578, %onnx::Conv_579)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_581, %onnx::Conv_582)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_584, %onnx::Conv_585)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_587, %onnx::Conv_588)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_590, %onnx::Conv_591)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_593, %onnx::Conv_594)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_596, %onnx::Conv_597)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_599, %onnx::Conv_600)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_602, %onnx::Conv_603)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_605, %onnx::Conv_606)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%567 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %567
}
|
val_accuracy
| 0 | 45,288,832 | 1,418,372 |
{'zcp_synflow': 68.34089013854131, 'zcp_zen': 59.33824920654297, 'zcp_epe_nas': 15.290078103969245, 'zcp_fisher': 0.04528410732746124, 'zcp_flops': 45288832.0, 'zcp_grad_norm': 16.135665893554688, 'zcp_grasp': -0.035559654235839844, 'zcp_jacov': -16.059811884343503, 'zcp_l2_norm': 531.6985473632812, 'zcp_nwot': 202.62408881389416, 'zcp_params': 1418372.0, 'zcp_plain': -0.0028612713795155287, 'zcp_snip': 27.129867553710938, 'lat_1080ti_1': 0.2295446402482167, 'lat_1080ti_32': 0.2149134327857213, 'lat_1080ti_64': 0.06550733035825941, 'lat_2080ti_1': 0.299726738941845, 'lat_2080ti_32': 0.19876257285357246, 'lat_2080ti_64': 0.09420915874515945, 'lat_essential_ph_1': 0.11320754716981132, 'lat_eyeriss': 0.14862560284598306, 'lat_fpga': 0.19755786223889013, 'lat_gold_6226': 0.2770052753039922, 'lat_gold_6240': 0.2950358408512843, 'lat_pixel2': 0.10869565217391304, 'lat_pixel3': 0.13817411849557767, 'lat_raspi4': 0.14642485609315667, 'lat_samsung_a50': 0.06315789473684211, 'lat_samsung_s7': 0.047244094488188976, 'lat_silver_4114': 0.3190642532462368, 'lat_silver_4210r': 0.29578932430143623, 'lat_titan_rtx_1': 0.29652310395942466, 'lat_titan_rtx_32': 0.22174284847130554, 'lat_titan_rtx_64': 0.10993119224249523, 'lat_titanx_1': 0.1662695283934922, 'lat_titanx_32': 0.13288508396616297, 'lat_titanx_64': 0.0670735050841498, 'lat_titanxp_1': 0.27606328414072934, 'lat_titanxp_32': 0.1818971535020431, 'lat_titanxp_64': 0.08898183815007389}
| |
FBNet_3325
|
FBNet
|
3325
|
3325
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_686[FLOAT, 16x3x3x3]
%onnx::Conv_687[FLOAT, 16]
%onnx::Conv_689[FLOAT, 48x16x1x1]
%onnx::Conv_690[FLOAT, 48]
%onnx::Conv_692[FLOAT, 48x1x5x5]
%onnx::Conv_695[FLOAT, 16x48x1x1]
%onnx::Conv_698[FLOAT, 96x16x1x1]
%onnx::Conv_699[FLOAT, 96]
%onnx::Conv_701[FLOAT, 96x1x3x3]
%onnx::Conv_704[FLOAT, 24x96x1x1]
%onnx::Conv_705[FLOAT, 24]
%onnx::Conv_707[FLOAT, 24x12x1x1]
%onnx::Conv_710[FLOAT, 24x1x5x5]
%onnx::Conv_713[FLOAT, 24x12x1x1]
%onnx::Conv_716[FLOAT, 72x24x1x1]
%onnx::Conv_717[FLOAT, 72]
%onnx::Conv_719[FLOAT, 72x1x3x3]
%onnx::Conv_722[FLOAT, 24x72x1x1]
%onnx::Conv_725[FLOAT, 24x24x1x1]
%onnx::Conv_728[FLOAT, 24x1x5x5]
%onnx::Conv_731[FLOAT, 24x24x1x1]
%onnx::Conv_734[FLOAT, 144x24x1x1]
%onnx::Conv_735[FLOAT, 144]
%onnx::Conv_737[FLOAT, 144x1x5x5]
%onnx::Conv_740[FLOAT, 32x144x1x1]
%onnx::Conv_741[FLOAT, 32]
%onnx::Conv_743[FLOAT, 192x32x1x1]
%onnx::Conv_744[FLOAT, 192]
%onnx::Conv_746[FLOAT, 192x1x3x3]
%onnx::Conv_749[FLOAT, 32x192x1x1]
%onnx::Conv_752[FLOAT, 192x32x1x1]
%onnx::Conv_755[FLOAT, 192x1x5x5]
%onnx::Conv_758[FLOAT, 32x192x1x1]
%onnx::Conv_761[FLOAT, 96x32x1x1]
%onnx::Conv_764[FLOAT, 96x1x5x5]
%onnx::Conv_767[FLOAT, 32x96x1x1]
%onnx::Conv_770[FLOAT, 32x32x1x1]
%onnx::Conv_773[FLOAT, 32x1x3x3]
%onnx::Conv_776[FLOAT, 64x32x1x1]
%onnx::Conv_777[FLOAT, 64]
%onnx::Conv_779[FLOAT, 384x64x1x1]
%onnx::Conv_780[FLOAT, 384]
%onnx::Conv_782[FLOAT, 384x1x3x3]
%onnx::Conv_785[FLOAT, 64x384x1x1]
%onnx::Conv_788[FLOAT, 64x64x1x1]
%onnx::Conv_791[FLOAT, 64x1x3x3]
%onnx::Conv_794[FLOAT, 64x64x1x1]
%onnx::Conv_797[FLOAT, 64x32x1x1]
%onnx::Conv_800[FLOAT, 64x1x3x3]
%onnx::Conv_803[FLOAT, 64x32x1x1]
%onnx::Conv_806[FLOAT, 384x64x1x1]
%onnx::Conv_809[FLOAT, 384x1x5x5]
%onnx::Conv_812[FLOAT, 112x384x1x1]
%onnx::Conv_813[FLOAT, 112]
%onnx::Conv_815[FLOAT, 112x56x1x1]
%onnx::Conv_818[FLOAT, 112x1x3x3]
%onnx::Conv_821[FLOAT, 112x56x1x1]
%onnx::Conv_824[FLOAT, 672x112x1x1]
%onnx::Conv_825[FLOAT, 672]
%onnx::Conv_827[FLOAT, 672x1x3x3]
%onnx::Conv_830[FLOAT, 112x672x1x1]
%onnx::Conv_833[FLOAT, 112x56x1x1]
%onnx::Conv_836[FLOAT, 112x1x5x5]
%onnx::Conv_839[FLOAT, 184x56x1x1]
%onnx::Conv_840[FLOAT, 184]
%onnx::Conv_842[FLOAT, 552x184x1x1]
%onnx::Conv_843[FLOAT, 552]
%onnx::Conv_845[FLOAT, 552x1x5x5]
%onnx::Conv_848[FLOAT, 184x552x1x1]
%onnx::Conv_851[FLOAT, 184x92x1x1]
%onnx::Conv_854[FLOAT, 184x1x5x5]
%onnx::Conv_857[FLOAT, 184x92x1x1]
%onnx::Conv_860[FLOAT, 1104x184x1x1]
%onnx::Conv_861[FLOAT, 1104]
%onnx::Conv_863[FLOAT, 1104x1x5x5]
%onnx::Conv_866[FLOAT, 184x1104x1x1]
%onnx::Conv_869[FLOAT, 1104x184x1x1]
%onnx::Conv_872[FLOAT, 1104x1x5x5]
%onnx::Conv_875[FLOAT, 352x1104x1x1]
%onnx::Conv_876[FLOAT, 352]
%onnx::Conv_878[FLOAT, 1504x352x1x1]
%onnx::Conv_879[FLOAT, 1504]
) {
%onnx::Conv_873 = Identity(%onnx::Conv_861)
%onnx::Conv_870 = Identity(%onnx::Conv_861)
%onnx::Conv_867 = Identity(%onnx::Conv_840)
%onnx::Conv_864 = Identity(%onnx::Conv_861)
%onnx::Conv_858 = Identity(%onnx::Conv_840)
%onnx::Conv_855 = Identity(%onnx::Conv_840)
%onnx::Conv_852 = Identity(%onnx::Conv_840)
%onnx::Conv_849 = Identity(%onnx::Conv_840)
%onnx::Conv_846 = Identity(%onnx::Conv_843)
%onnx::Conv_837 = Identity(%onnx::Conv_813)
%onnx::Conv_834 = Identity(%onnx::Conv_813)
%onnx::Conv_831 = Identity(%onnx::Conv_813)
%onnx::Conv_828 = Identity(%onnx::Conv_825)
%onnx::Conv_822 = Identity(%onnx::Conv_813)
%onnx::Conv_819 = Identity(%onnx::Conv_813)
%onnx::Conv_816 = Identity(%onnx::Conv_813)
%onnx::Conv_810 = Identity(%onnx::Conv_780)
%onnx::Conv_807 = Identity(%onnx::Conv_780)
%onnx::Conv_804 = Identity(%onnx::Conv_777)
%onnx::Conv_801 = Identity(%onnx::Conv_777)
%onnx::Conv_798 = Identity(%onnx::Conv_777)
%onnx::Conv_795 = Identity(%onnx::Conv_777)
%onnx::Conv_792 = Identity(%onnx::Conv_777)
%onnx::Conv_789 = Identity(%onnx::Conv_777)
%onnx::Conv_786 = Identity(%onnx::Conv_777)
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_774 = Identity(%onnx::Conv_741)
%onnx::Conv_771 = Identity(%onnx::Conv_741)
%onnx::Conv_768 = Identity(%onnx::Conv_741)
%onnx::Conv_765 = Identity(%onnx::Conv_699)
%onnx::Conv_762 = Identity(%onnx::Conv_699)
%onnx::Conv_759 = Identity(%onnx::Conv_741)
%onnx::Conv_756 = Identity(%onnx::Conv_744)
%onnx::Conv_753 = Identity(%onnx::Conv_744)
%onnx::Conv_750 = Identity(%onnx::Conv_741)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_732 = Identity(%onnx::Conv_705)
%onnx::Conv_729 = Identity(%onnx::Conv_705)
%onnx::Conv_726 = Identity(%onnx::Conv_705)
%onnx::Conv_723 = Identity(%onnx::Conv_705)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%onnx::Conv_714 = Identity(%onnx::Conv_705)
%onnx::Conv_711 = Identity(%onnx::Conv_705)
%onnx::Conv_708 = Identity(%onnx::Conv_705)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%onnx::Conv_696 = Identity(%onnx::Conv_687)
%onnx::Conv_693 = Identity(%onnx::Conv_690)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_686, %onnx::Conv_687)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_872, %onnx::Conv_873)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_875, %onnx::Conv_876)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_878, %onnx::Conv_879)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%684 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %684
}
|
val_accuracy
| 0 | 85,426,048 | 2,420,332 |
{'zcp_synflow': 79.06249648843257, 'zcp_zen': 72.06387329101562, 'zcp_epe_nas': 7.690549507757447, 'zcp_fisher': 0.1706494241952896, 'zcp_flops': 85426048.0, 'zcp_grad_norm': 29.44540786743164, 'zcp_grasp': -0.14976119995117188, 'zcp_jacov': -16.05869445766561, 'zcp_l2_norm': 674.9317626953125, 'zcp_nwot': 216.27554054661354, 'zcp_params': 2420332.0, 'zcp_plain': -0.003205366898328066, 'zcp_snip': 52.10393142700195, 'lat_1080ti_1': 0.6895691152492889, 'lat_1080ti_32': 0.6153273461919148, 'lat_1080ti_64': 0.602052535568125, 'lat_2080ti_1': 0.7265347445927357, 'lat_2080ti_32': 0.6189327818603089, 'lat_2080ti_64': 0.6081834633974116, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.6728127922542315, 'lat_fpga': 0.6965534984513998, 'lat_gold_6226': 0.727030849887149, 'lat_gold_6240': 0.707877402027253, 'lat_pixel2': 0.45652173913043476, 'lat_pixel3': 0.6445061263319399, 'lat_raspi4': 0.7051628073661128, 'lat_samsung_a50': 0.29473684210526313, 'lat_samsung_s7': 0.25196850393700787, 'lat_silver_4114': 0.7418262161867225, 'lat_silver_4210r': 0.7980013347750181, 'lat_titan_rtx_1': 0.700301047092735, 'lat_titan_rtx_32': 0.592531647960304, 'lat_titan_rtx_64': 0.6116507573452482, 'lat_titanx_1': 0.3731797787250534, 'lat_titanx_32': 0.6012654109245548, 'lat_titanx_64': 0.6523375614068149, 'lat_titanxp_1': 0.6748607812383498, 'lat_titanxp_32': 0.604454892565926, 'lat_titanxp_64': 0.6019000125453103}
| |
FBNet_185
|
FBNet
|
185
|
185
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_634[FLOAT, 16x3x3x3]
%onnx::Conv_635[FLOAT, 16]
%onnx::Conv_637[FLOAT, 16x16x1x1]
%onnx::Conv_640[FLOAT, 16x1x5x5]
%onnx::Conv_643[FLOAT, 16x16x1x1]
%onnx::Conv_646[FLOAT, 48x16x1x1]
%onnx::Conv_647[FLOAT, 48]
%onnx::Conv_649[FLOAT, 48x1x5x5]
%onnx::Conv_652[FLOAT, 24x48x1x1]
%onnx::Conv_653[FLOAT, 24]
%onnx::Conv_655[FLOAT, 72x24x1x1]
%onnx::Conv_656[FLOAT, 72]
%onnx::Conv_658[FLOAT, 72x1x3x3]
%onnx::Conv_661[FLOAT, 24x72x1x1]
%onnx::Conv_664[FLOAT, 144x24x1x1]
%onnx::Conv_665[FLOAT, 144]
%onnx::Conv_667[FLOAT, 144x1x3x3]
%onnx::Conv_670[FLOAT, 24x144x1x1]
%onnx::Conv_673[FLOAT, 24x12x1x1]
%onnx::Conv_676[FLOAT, 24x1x5x5]
%onnx::Conv_679[FLOAT, 32x12x1x1]
%onnx::Conv_680[FLOAT, 32]
%onnx::Conv_682[FLOAT, 96x32x1x1]
%onnx::Conv_683[FLOAT, 96]
%onnx::Conv_685[FLOAT, 96x1x3x3]
%onnx::Conv_688[FLOAT, 32x96x1x1]
%onnx::Conv_691[FLOAT, 32x16x1x1]
%onnx::Conv_694[FLOAT, 32x1x3x3]
%onnx::Conv_697[FLOAT, 32x16x1x1]
%onnx::Conv_700[FLOAT, 32x32x1x1]
%onnx::Conv_703[FLOAT, 32x1x5x5]
%onnx::Conv_706[FLOAT, 32x32x1x1]
%onnx::Conv_709[FLOAT, 192x32x1x1]
%onnx::Conv_710[FLOAT, 192]
%onnx::Conv_712[FLOAT, 192x1x3x3]
%onnx::Conv_715[FLOAT, 64x192x1x1]
%onnx::Conv_716[FLOAT, 64]
%onnx::Conv_718[FLOAT, 192x64x1x1]
%onnx::Conv_721[FLOAT, 192x1x3x3]
%onnx::Conv_724[FLOAT, 64x192x1x1]
%onnx::Conv_727[FLOAT, 384x64x1x1]
%onnx::Conv_728[FLOAT, 384]
%onnx::Conv_730[FLOAT, 384x1x3x3]
%onnx::Conv_733[FLOAT, 64x384x1x1]
%onnx::Conv_736[FLOAT, 64x32x1x1]
%onnx::Conv_739[FLOAT, 64x1x5x5]
%onnx::Conv_742[FLOAT, 112x32x1x1]
%onnx::Conv_743[FLOAT, 112]
%onnx::Conv_745[FLOAT, 112x56x1x1]
%onnx::Conv_748[FLOAT, 112x1x5x5]
%onnx::Conv_751[FLOAT, 112x56x1x1]
%onnx::Conv_754[FLOAT, 112x56x1x1]
%onnx::Conv_757[FLOAT, 112x1x3x3]
%onnx::Conv_760[FLOAT, 112x56x1x1]
%onnx::Conv_763[FLOAT, 336x112x1x1]
%onnx::Conv_764[FLOAT, 336]
%onnx::Conv_766[FLOAT, 336x1x3x3]
%onnx::Conv_769[FLOAT, 112x336x1x1]
%onnx::Conv_772[FLOAT, 672x112x1x1]
%onnx::Conv_773[FLOAT, 672]
%onnx::Conv_775[FLOAT, 672x1x5x5]
%onnx::Conv_778[FLOAT, 184x672x1x1]
%onnx::Conv_779[FLOAT, 184]
%onnx::Conv_781[FLOAT, 184x92x1x1]
%onnx::Conv_784[FLOAT, 184x1x3x3]
%onnx::Conv_787[FLOAT, 184x92x1x1]
%onnx::Conv_790[FLOAT, 1104x184x1x1]
%onnx::Conv_791[FLOAT, 1104]
%onnx::Conv_793[FLOAT, 1104x1x5x5]
%onnx::Conv_796[FLOAT, 184x1104x1x1]
%onnx::Conv_799[FLOAT, 352x184x1x1]
%onnx::Conv_800[FLOAT, 352]
%onnx::Conv_802[FLOAT, 1504x352x1x1]
%onnx::Conv_803[FLOAT, 1504]
) {
%onnx::Conv_797 = Identity(%onnx::Conv_779)
%onnx::Conv_794 = Identity(%onnx::Conv_791)
%onnx::Conv_788 = Identity(%onnx::Conv_779)
%onnx::Conv_785 = Identity(%onnx::Conv_779)
%onnx::Conv_782 = Identity(%onnx::Conv_779)
%onnx::Conv_776 = Identity(%onnx::Conv_773)
%onnx::Conv_770 = Identity(%onnx::Conv_743)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_761 = Identity(%onnx::Conv_743)
%onnx::Conv_758 = Identity(%onnx::Conv_743)
%onnx::Conv_755 = Identity(%onnx::Conv_743)
%onnx::Conv_752 = Identity(%onnx::Conv_743)
%onnx::Conv_749 = Identity(%onnx::Conv_743)
%onnx::Conv_746 = Identity(%onnx::Conv_743)
%onnx::Conv_740 = Identity(%onnx::Conv_716)
%onnx::Conv_737 = Identity(%onnx::Conv_716)
%onnx::Conv_734 = Identity(%onnx::Conv_716)
%onnx::Conv_731 = Identity(%onnx::Conv_728)
%onnx::Conv_725 = Identity(%onnx::Conv_716)
%onnx::Conv_722 = Identity(%onnx::Conv_710)
%onnx::Conv_719 = Identity(%onnx::Conv_710)
%onnx::Conv_713 = Identity(%onnx::Conv_710)
%onnx::Conv_707 = Identity(%onnx::Conv_680)
%onnx::Conv_704 = Identity(%onnx::Conv_680)
%onnx::Conv_701 = Identity(%onnx::Conv_680)
%onnx::Conv_698 = Identity(%onnx::Conv_680)
%onnx::Conv_695 = Identity(%onnx::Conv_680)
%onnx::Conv_692 = Identity(%onnx::Conv_680)
%onnx::Conv_689 = Identity(%onnx::Conv_680)
%onnx::Conv_686 = Identity(%onnx::Conv_683)
%onnx::Conv_677 = Identity(%onnx::Conv_653)
%onnx::Conv_674 = Identity(%onnx::Conv_653)
%onnx::Conv_671 = Identity(%onnx::Conv_653)
%onnx::Conv_668 = Identity(%onnx::Conv_665)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_656)
%onnx::Conv_650 = Identity(%onnx::Conv_647)
%onnx::Conv_644 = Identity(%onnx::Conv_635)
%onnx::Conv_641 = Identity(%onnx::Conv_635)
%onnx::Conv_638 = Identity(%onnx::Conv_635)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_634, %onnx::Conv_635)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%632 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %632
}
|
val_accuracy
| 0 | 62,612,864 | 1,694,300 |
{'zcp_synflow': 68.74602816355535, 'zcp_zen': 60.45960235595703, 'zcp_epe_nas': 9.649076676271427, 'zcp_fisher': 0.08040807396173477, 'zcp_flops': 62612864.0, 'zcp_grad_norm': 19.699432373046875, 'zcp_grasp': 0.0004596710205078125, 'zcp_jacov': -16.03883803626119, 'zcp_l2_norm': 544.441650390625, 'zcp_nwot': 211.9802238936396, 'zcp_params': 1694300.0, 'zcp_plain': 0.004604548215866089, 'zcp_snip': 32.15994644165039, 'lat_1080ti_1': 0.48570397036482493, 'lat_1080ti_32': 0.39074894737461674, 'lat_1080ti_64': 0.35698566595562026, 'lat_2080ti_1': 0.4506378266996825, 'lat_2080ti_32': 0.42076099018005675, 'lat_2080ti_64': 0.37240385409212723, 'lat_essential_ph_1': 0.37735849056603776, 'lat_eyeriss': 0.3638347422661508, 'lat_fpga': 0.35238820886808675, 'lat_gold_6226': 0.2958125986973109, 'lat_gold_6240': 0.4521278981693365, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.33582261816974107, 'lat_raspi4': 0.35973109127259006, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.1889763779527559, 'lat_silver_4114': 0.41375913539102566, 'lat_silver_4210r': 0.4066065289960339, 'lat_titan_rtx_1': 0.4409590020010048, 'lat_titan_rtx_32': 0.41435126380270676, 'lat_titan_rtx_64': 0.39326204004404686, 'lat_titanx_1': 0.24574168104570812, 'lat_titanx_32': 0.3874856740671061, 'lat_titanx_64': 0.35673997658476186, 'lat_titanxp_1': 0.4127718073837936, 'lat_titanxp_32': 0.40877379049329854, 'lat_titanxp_64': 0.36778580419849277}
| |
FBNet_3133
|
FBNet
|
3133
|
3133
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_678[FLOAT, 16x3x3x3]
%onnx::Conv_679[FLOAT, 16]
%onnx::Conv_681[FLOAT, 16x16x1x1]
%onnx::Conv_684[FLOAT, 16x1x3x3]
%onnx::Conv_687[FLOAT, 16x16x1x1]
%onnx::Conv_690[FLOAT, 48x16x1x1]
%onnx::Conv_691[FLOAT, 48]
%onnx::Conv_693[FLOAT, 48x1x5x5]
%onnx::Conv_696[FLOAT, 24x48x1x1]
%onnx::Conv_697[FLOAT, 24]
%onnx::Conv_699[FLOAT, 72x24x1x1]
%onnx::Conv_700[FLOAT, 72]
%onnx::Conv_702[FLOAT, 72x1x5x5]
%onnx::Conv_705[FLOAT, 24x72x1x1]
%onnx::Conv_708[FLOAT, 24x24x1x1]
%onnx::Conv_711[FLOAT, 24x1x3x3]
%onnx::Conv_714[FLOAT, 24x24x1x1]
%onnx::Conv_717[FLOAT, 144x24x1x1]
%onnx::Conv_718[FLOAT, 144]
%onnx::Conv_720[FLOAT, 144x1x3x3]
%onnx::Conv_723[FLOAT, 24x144x1x1]
%onnx::Conv_726[FLOAT, 24x24x1x1]
%onnx::Conv_729[FLOAT, 24x1x3x3]
%onnx::Conv_732[FLOAT, 32x24x1x1]
%onnx::Conv_733[FLOAT, 32]
%onnx::Conv_735[FLOAT, 32x16x1x1]
%onnx::Conv_738[FLOAT, 32x1x3x3]
%onnx::Conv_741[FLOAT, 32x16x1x1]
%onnx::Conv_744[FLOAT, 32x32x1x1]
%onnx::Conv_747[FLOAT, 32x1x3x3]
%onnx::Conv_750[FLOAT, 32x32x1x1]
%onnx::Conv_753[FLOAT, 192x32x1x1]
%onnx::Conv_754[FLOAT, 192]
%onnx::Conv_756[FLOAT, 192x1x5x5]
%onnx::Conv_759[FLOAT, 32x192x1x1]
%onnx::Conv_762[FLOAT, 32x32x1x1]
%onnx::Conv_765[FLOAT, 32x1x5x5]
%onnx::Conv_768[FLOAT, 64x32x1x1]
%onnx::Conv_769[FLOAT, 64]
%onnx::Conv_771[FLOAT, 192x64x1x1]
%onnx::Conv_774[FLOAT, 192x1x5x5]
%onnx::Conv_777[FLOAT, 64x192x1x1]
%onnx::Conv_780[FLOAT, 64x64x1x1]
%onnx::Conv_783[FLOAT, 64x1x3x3]
%onnx::Conv_786[FLOAT, 64x64x1x1]
%onnx::Conv_789[FLOAT, 64x32x1x1]
%onnx::Conv_792[FLOAT, 64x1x3x3]
%onnx::Conv_795[FLOAT, 64x32x1x1]
%onnx::Conv_798[FLOAT, 64x64x1x1]
%onnx::Conv_801[FLOAT, 64x1x5x5]
%onnx::Conv_804[FLOAT, 112x64x1x1]
%onnx::Conv_805[FLOAT, 112]
%onnx::Conv_807[FLOAT, 672x112x1x1]
%onnx::Conv_808[FLOAT, 672]
%onnx::Conv_810[FLOAT, 672x1x3x3]
%onnx::Conv_813[FLOAT, 112x672x1x1]
%onnx::Conv_816[FLOAT, 112x112x1x1]
%onnx::Conv_819[FLOAT, 112x1x3x3]
%onnx::Conv_822[FLOAT, 112x112x1x1]
%onnx::Conv_825[FLOAT, 112x56x1x1]
%onnx::Conv_828[FLOAT, 112x1x3x3]
%onnx::Conv_831[FLOAT, 112x56x1x1]
%onnx::Conv_834[FLOAT, 112x112x1x1]
%onnx::Conv_837[FLOAT, 112x1x3x3]
%onnx::Conv_840[FLOAT, 184x112x1x1]
%onnx::Conv_841[FLOAT, 184]
%onnx::Conv_843[FLOAT, 552x184x1x1]
%onnx::Conv_844[FLOAT, 552]
%onnx::Conv_846[FLOAT, 552x1x5x5]
%onnx::Conv_849[FLOAT, 184x552x1x1]
%onnx::Conv_852[FLOAT, 184x92x1x1]
%onnx::Conv_855[FLOAT, 184x1x5x5]
%onnx::Conv_858[FLOAT, 184x92x1x1]
%onnx::Conv_861[FLOAT, 1104x184x1x1]
%onnx::Conv_862[FLOAT, 1104]
%onnx::Conv_864[FLOAT, 1104x1x5x5]
%onnx::Conv_867[FLOAT, 184x1104x1x1]
%onnx::Conv_870[FLOAT, 352x184x1x1]
%onnx::Conv_871[FLOAT, 352]
%onnx::Conv_873[FLOAT, 1504x352x1x1]
%onnx::Conv_874[FLOAT, 1504]
) {
%onnx::Conv_868 = Identity(%onnx::Conv_841)
%onnx::Conv_865 = Identity(%onnx::Conv_862)
%onnx::Conv_859 = Identity(%onnx::Conv_841)
%onnx::Conv_856 = Identity(%onnx::Conv_841)
%onnx::Conv_853 = Identity(%onnx::Conv_841)
%onnx::Conv_850 = Identity(%onnx::Conv_841)
%onnx::Conv_847 = Identity(%onnx::Conv_844)
%onnx::Conv_838 = Identity(%onnx::Conv_805)
%onnx::Conv_835 = Identity(%onnx::Conv_805)
%onnx::Conv_832 = Identity(%onnx::Conv_805)
%onnx::Conv_829 = Identity(%onnx::Conv_805)
%onnx::Conv_826 = Identity(%onnx::Conv_805)
%onnx::Conv_823 = Identity(%onnx::Conv_805)
%onnx::Conv_820 = Identity(%onnx::Conv_805)
%onnx::Conv_817 = Identity(%onnx::Conv_805)
%onnx::Conv_814 = Identity(%onnx::Conv_805)
%onnx::Conv_811 = Identity(%onnx::Conv_808)
%onnx::Conv_802 = Identity(%onnx::Conv_769)
%onnx::Conv_799 = Identity(%onnx::Conv_769)
%onnx::Conv_796 = Identity(%onnx::Conv_769)
%onnx::Conv_793 = Identity(%onnx::Conv_769)
%onnx::Conv_790 = Identity(%onnx::Conv_769)
%onnx::Conv_787 = Identity(%onnx::Conv_769)
%onnx::Conv_784 = Identity(%onnx::Conv_769)
%onnx::Conv_781 = Identity(%onnx::Conv_769)
%onnx::Conv_778 = Identity(%onnx::Conv_769)
%onnx::Conv_775 = Identity(%onnx::Conv_754)
%onnx::Conv_772 = Identity(%onnx::Conv_754)
%onnx::Conv_766 = Identity(%onnx::Conv_733)
%onnx::Conv_763 = Identity(%onnx::Conv_733)
%onnx::Conv_760 = Identity(%onnx::Conv_733)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_751 = Identity(%onnx::Conv_733)
%onnx::Conv_748 = Identity(%onnx::Conv_733)
%onnx::Conv_745 = Identity(%onnx::Conv_733)
%onnx::Conv_742 = Identity(%onnx::Conv_733)
%onnx::Conv_739 = Identity(%onnx::Conv_733)
%onnx::Conv_736 = Identity(%onnx::Conv_733)
%onnx::Conv_730 = Identity(%onnx::Conv_697)
%onnx::Conv_727 = Identity(%onnx::Conv_697)
%onnx::Conv_724 = Identity(%onnx::Conv_697)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%onnx::Conv_715 = Identity(%onnx::Conv_697)
%onnx::Conv_712 = Identity(%onnx::Conv_697)
%onnx::Conv_709 = Identity(%onnx::Conv_697)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_700)
%onnx::Conv_694 = Identity(%onnx::Conv_691)
%onnx::Conv_688 = Identity(%onnx::Conv_679)
%onnx::Conv_685 = Identity(%onnx::Conv_679)
%onnx::Conv_682 = Identity(%onnx::Conv_679)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_678, %onnx::Conv_679)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%676 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %676
}
|
val_accuracy
| 0 | 67,169,152 | 1,787,708 |
{'zcp_synflow': 81.96152541839068, 'zcp_zen': 69.47039794921875, 'zcp_epe_nas': 8.437835874849965, 'zcp_fisher': 0.18082435429096222, 'zcp_flops': 67169152.0, 'zcp_grad_norm': 22.129894256591797, 'zcp_grasp': -0.10228729248046875, 'zcp_jacov': -16.061361159590405, 'zcp_l2_norm': 617.6553955078125, 'zcp_nwot': 212.88566097843102, 'zcp_params': 1787708.0, 'zcp_plain': -0.000570389092899859, 'zcp_snip': 39.69680404663086, 'lat_1080ti_1': 0.7221200517841212, 'lat_1080ti_32': 0.6211923030904024, 'lat_1080ti_64': 0.5559954401145112, 'lat_2080ti_1': 0.7585488473287941, 'lat_2080ti_32': 0.6659850150723334, 'lat_2080ti_64': 0.5840693042168889, 'lat_essential_ph_1': 0.3018867924528302, 'lat_eyeriss': 0.4387435131021585, 'lat_fpga': 0.4529829109727979, 'lat_gold_6226': 0.303030512267516, 'lat_gold_6240': 0.4781857455857433, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.4181505248530033, 'lat_raspi4': 0.44124357977280093, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.16535433070866143, 'lat_silver_4114': 0.5198021663787378, 'lat_silver_4210r': 0.511472965678492, 'lat_titan_rtx_1': 0.705573676609371, 'lat_titan_rtx_32': 0.6393878629718973, 'lat_titan_rtx_64': 0.6115226976245001, 'lat_titanx_1': 0.37559354139893375, 'lat_titanx_32': 0.6373639425218272, 'lat_titanx_64': 0.5181140765612401, 'lat_titanxp_1': 0.66622250758849, 'lat_titanxp_32': 0.6639103362299056, 'lat_titanxp_64': 0.582600878421106}
| |
FBNet_4887
|
FBNet
|
4887
|
4887
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_634[FLOAT, 16x3x3x3]
%onnx::Conv_635[FLOAT, 16]
%onnx::Conv_637[FLOAT, 96x16x1x1]
%onnx::Conv_638[FLOAT, 96]
%onnx::Conv_640[FLOAT, 96x1x3x3]
%onnx::Conv_643[FLOAT, 16x96x1x1]
%onnx::Conv_646[FLOAT, 16x8x1x1]
%onnx::Conv_649[FLOAT, 16x1x5x5]
%onnx::Conv_652[FLOAT, 24x8x1x1]
%onnx::Conv_653[FLOAT, 24]
%onnx::Conv_655[FLOAT, 24x24x1x1]
%onnx::Conv_658[FLOAT, 24x1x3x3]
%onnx::Conv_661[FLOAT, 24x24x1x1]
%onnx::Conv_664[FLOAT, 24x12x1x1]
%onnx::Conv_667[FLOAT, 24x1x3x3]
%onnx::Conv_670[FLOAT, 24x12x1x1]
%onnx::Conv_673[FLOAT, 24x12x1x1]
%onnx::Conv_676[FLOAT, 24x1x3x3]
%onnx::Conv_679[FLOAT, 24x12x1x1]
%onnx::Conv_682[FLOAT, 32x24x1x1]
%onnx::Conv_683[FLOAT, 32]
%onnx::Conv_685[FLOAT, 192x32x1x1]
%onnx::Conv_686[FLOAT, 192]
%onnx::Conv_688[FLOAT, 192x1x3x3]
%onnx::Conv_691[FLOAT, 32x192x1x1]
%onnx::Conv_694[FLOAT, 192x32x1x1]
%onnx::Conv_697[FLOAT, 192x1x5x5]
%onnx::Conv_700[FLOAT, 32x192x1x1]
%onnx::Conv_703[FLOAT, 32x16x1x1]
%onnx::Conv_706[FLOAT, 32x1x5x5]
%onnx::Conv_709[FLOAT, 32x16x1x1]
%onnx::Conv_712[FLOAT, 32x32x1x1]
%onnx::Conv_715[FLOAT, 32x1x5x5]
%onnx::Conv_718[FLOAT, 64x32x1x1]
%onnx::Conv_719[FLOAT, 64]
%onnx::Conv_721[FLOAT, 64x32x1x1]
%onnx::Conv_724[FLOAT, 64x1x3x3]
%onnx::Conv_727[FLOAT, 64x32x1x1]
%onnx::Conv_730[FLOAT, 192x64x1x1]
%onnx::Conv_733[FLOAT, 192x1x5x5]
%onnx::Conv_736[FLOAT, 64x192x1x1]
%onnx::Conv_739[FLOAT, 64x64x1x1]
%onnx::Conv_742[FLOAT, 64x1x5x5]
%onnx::Conv_745[FLOAT, 112x64x1x1]
%onnx::Conv_746[FLOAT, 112]
%onnx::Conv_748[FLOAT, 112x112x1x1]
%onnx::Conv_751[FLOAT, 112x1x3x3]
%onnx::Conv_754[FLOAT, 112x112x1x1]
%onnx::Conv_757[FLOAT, 336x112x1x1]
%onnx::Conv_758[FLOAT, 336]
%onnx::Conv_760[FLOAT, 336x1x5x5]
%onnx::Conv_763[FLOAT, 184x336x1x1]
%onnx::Conv_764[FLOAT, 184]
%onnx::Conv_766[FLOAT, 184x92x1x1]
%onnx::Conv_769[FLOAT, 184x1x3x3]
%onnx::Conv_772[FLOAT, 184x92x1x1]
%onnx::Conv_775[FLOAT, 1104x184x1x1]
%onnx::Conv_776[FLOAT, 1104]
%onnx::Conv_778[FLOAT, 1104x1x3x3]
%onnx::Conv_781[FLOAT, 184x1104x1x1]
%onnx::Conv_784[FLOAT, 1104x184x1x1]
%onnx::Conv_787[FLOAT, 1104x1x5x5]
%onnx::Conv_790[FLOAT, 184x1104x1x1]
%onnx::Conv_793[FLOAT, 552x184x1x1]
%onnx::Conv_794[FLOAT, 552]
%onnx::Conv_796[FLOAT, 552x1x5x5]
%onnx::Conv_799[FLOAT, 352x552x1x1]
%onnx::Conv_800[FLOAT, 352]
%onnx::Conv_802[FLOAT, 1504x352x1x1]
%onnx::Conv_803[FLOAT, 1504]
) {
%onnx::Conv_797 = Identity(%onnx::Conv_794)
%onnx::Conv_791 = Identity(%onnx::Conv_764)
%onnx::Conv_788 = Identity(%onnx::Conv_776)
%onnx::Conv_785 = Identity(%onnx::Conv_776)
%onnx::Conv_782 = Identity(%onnx::Conv_764)
%onnx::Conv_779 = Identity(%onnx::Conv_776)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_764)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_761 = Identity(%onnx::Conv_758)
%onnx::Conv_755 = Identity(%onnx::Conv_746)
%onnx::Conv_752 = Identity(%onnx::Conv_746)
%onnx::Conv_749 = Identity(%onnx::Conv_746)
%onnx::Conv_743 = Identity(%onnx::Conv_719)
%onnx::Conv_740 = Identity(%onnx::Conv_719)
%onnx::Conv_737 = Identity(%onnx::Conv_719)
%onnx::Conv_734 = Identity(%onnx::Conv_686)
%onnx::Conv_731 = Identity(%onnx::Conv_686)
%onnx::Conv_728 = Identity(%onnx::Conv_719)
%onnx::Conv_725 = Identity(%onnx::Conv_719)
%onnx::Conv_722 = Identity(%onnx::Conv_719)
%onnx::Conv_716 = Identity(%onnx::Conv_683)
%onnx::Conv_713 = Identity(%onnx::Conv_683)
%onnx::Conv_710 = Identity(%onnx::Conv_683)
%onnx::Conv_707 = Identity(%onnx::Conv_683)
%onnx::Conv_704 = Identity(%onnx::Conv_683)
%onnx::Conv_701 = Identity(%onnx::Conv_683)
%onnx::Conv_698 = Identity(%onnx::Conv_686)
%onnx::Conv_695 = Identity(%onnx::Conv_686)
%onnx::Conv_692 = Identity(%onnx::Conv_683)
%onnx::Conv_689 = Identity(%onnx::Conv_686)
%onnx::Conv_680 = Identity(%onnx::Conv_653)
%onnx::Conv_677 = Identity(%onnx::Conv_653)
%onnx::Conv_674 = Identity(%onnx::Conv_653)
%onnx::Conv_671 = Identity(%onnx::Conv_653)
%onnx::Conv_668 = Identity(%onnx::Conv_653)
%onnx::Conv_665 = Identity(%onnx::Conv_653)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_650 = Identity(%onnx::Conv_635)
%onnx::Conv_647 = Identity(%onnx::Conv_635)
%onnx::Conv_644 = Identity(%onnx::Conv_635)
%onnx::Conv_641 = Identity(%onnx::Conv_638)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_634, %onnx::Conv_635)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.4/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%632 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %632
}
|
val_accuracy
| 0 | 55,719,808 | 2,125,548 |
{'zcp_synflow': 68.62744326553164, 'zcp_zen': 59.72840118408203, 'zcp_epe_nas': 7.351079226344437, 'zcp_fisher': 0.09177848696708679, 'zcp_flops': 55719808.0, 'zcp_grad_norm': 20.261268615722656, 'zcp_grasp': -0.3230857849121094, 'zcp_jacov': -16.06065659677271, 'zcp_l2_norm': 555.0804443359375, 'zcp_nwot': 208.4214928824, 'zcp_params': 2125548.0, 'zcp_plain': -0.003497544676065445, 'zcp_snip': 38.21852111816406, 'lat_1080ti_1': 0.5028813582505955, 'lat_1080ti_32': 0.38628802708976223, 'lat_1080ti_64': 0.253527736977345, 'lat_2080ti_1': 0.494932712828738, 'lat_2080ti_32': 0.39242023681168814, 'lat_2080ti_64': 0.3099896170660354, 'lat_essential_ph_1': 0.1320754716981132, 'lat_eyeriss': 0.34853208149194065, 'lat_fpga': 0.330440633689231, 'lat_gold_6226': 0.3559978642123364, 'lat_gold_6240': 0.46936598475575286, 'lat_pixel2': 0.391304347826087, 'lat_pixel3': 0.31828300013470645, 'lat_raspi4': 0.4224786811195975, 'lat_samsung_a50': 0.15789473684210525, 'lat_samsung_s7': 0.14173228346456693, 'lat_silver_4114': 0.5814547768176744, 'lat_silver_4210r': 0.47180483075592694, 'lat_titan_rtx_1': 0.459942028546155, 'lat_titan_rtx_32': 0.3723324818024436, 'lat_titan_rtx_64': 0.3109530494657894, 'lat_titanx_1': 0.2440177599378782, 'lat_titanx_32': 0.3086699609326806, 'lat_titanx_64': 0.26144943740932286, 'lat_titanxp_1': 0.4605906058269278, 'lat_titanxp_32': 0.3419709925226744, 'lat_titanxp_64': 0.27551575293068825}
| |
FBNet_3119
|
FBNet
|
3119
|
3119
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_588[FLOAT, 16x3x3x3]
%onnx::Conv_589[FLOAT, 16]
%onnx::Conv_591[FLOAT, 16x8x1x1]
%onnx::Conv_594[FLOAT, 16x1x5x5]
%onnx::Conv_597[FLOAT, 16x8x1x1]
%onnx::Conv_600[FLOAT, 48x16x1x1]
%onnx::Conv_601[FLOAT, 48]
%onnx::Conv_603[FLOAT, 48x1x5x5]
%onnx::Conv_606[FLOAT, 24x48x1x1]
%onnx::Conv_607[FLOAT, 24]
%onnx::Conv_609[FLOAT, 144x24x1x1]
%onnx::Conv_610[FLOAT, 144]
%onnx::Conv_612[FLOAT, 144x1x5x5]
%onnx::Conv_615[FLOAT, 24x144x1x1]
%onnx::Conv_618[FLOAT, 24x12x1x1]
%onnx::Conv_621[FLOAT, 24x1x5x5]
%onnx::Conv_624[FLOAT, 24x12x1x1]
%onnx::Conv_627[FLOAT, 72x24x1x1]
%onnx::Conv_628[FLOAT, 72]
%onnx::Conv_630[FLOAT, 72x1x3x3]
%onnx::Conv_633[FLOAT, 24x72x1x1]
%onnx::Conv_636[FLOAT, 24x12x1x1]
%onnx::Conv_639[FLOAT, 24x1x3x3]
%onnx::Conv_642[FLOAT, 32x12x1x1]
%onnx::Conv_643[FLOAT, 32]
%onnx::Conv_645[FLOAT, 32x32x1x1]
%onnx::Conv_648[FLOAT, 32x1x5x5]
%onnx::Conv_651[FLOAT, 32x32x1x1]
%onnx::Conv_654[FLOAT, 96x32x1x1]
%onnx::Conv_655[FLOAT, 96]
%onnx::Conv_657[FLOAT, 96x1x3x3]
%onnx::Conv_660[FLOAT, 64x96x1x1]
%onnx::Conv_661[FLOAT, 64]
%onnx::Conv_663[FLOAT, 64x64x1x1]
%onnx::Conv_666[FLOAT, 64x1x3x3]
%onnx::Conv_669[FLOAT, 64x64x1x1]
%onnx::Conv_672[FLOAT, 64x32x1x1]
%onnx::Conv_675[FLOAT, 64x1x5x5]
%onnx::Conv_678[FLOAT, 64x32x1x1]
%onnx::Conv_681[FLOAT, 192x64x1x1]
%onnx::Conv_682[FLOAT, 192]
%onnx::Conv_684[FLOAT, 192x1x3x3]
%onnx::Conv_687[FLOAT, 64x192x1x1]
%onnx::Conv_690[FLOAT, 384x64x1x1]
%onnx::Conv_691[FLOAT, 384]
%onnx::Conv_693[FLOAT, 384x1x3x3]
%onnx::Conv_696[FLOAT, 112x384x1x1]
%onnx::Conv_697[FLOAT, 112]
%onnx::Conv_699[FLOAT, 112x112x1x1]
%onnx::Conv_702[FLOAT, 112x1x3x3]
%onnx::Conv_705[FLOAT, 112x112x1x1]
%onnx::Conv_708[FLOAT, 112x56x1x1]
%onnx::Conv_711[FLOAT, 112x1x5x5]
%onnx::Conv_714[FLOAT, 112x56x1x1]
%onnx::Conv_717[FLOAT, 672x112x1x1]
%onnx::Conv_718[FLOAT, 672]
%onnx::Conv_720[FLOAT, 672x1x5x5]
%onnx::Conv_723[FLOAT, 184x672x1x1]
%onnx::Conv_724[FLOAT, 184]
%onnx::Conv_726[FLOAT, 184x184x1x1]
%onnx::Conv_729[FLOAT, 184x1x3x3]
%onnx::Conv_732[FLOAT, 184x184x1x1]
%onnx::Conv_735[FLOAT, 1104x184x1x1]
%onnx::Conv_736[FLOAT, 1104]
%onnx::Conv_738[FLOAT, 1104x1x5x5]
%onnx::Conv_741[FLOAT, 184x1104x1x1]
%onnx::Conv_744[FLOAT, 352x184x1x1]
%onnx::Conv_745[FLOAT, 352]
%onnx::Conv_747[FLOAT, 1504x352x1x1]
%onnx::Conv_748[FLOAT, 1504]
) {
%onnx::Conv_742 = Identity(%onnx::Conv_724)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_733 = Identity(%onnx::Conv_724)
%onnx::Conv_730 = Identity(%onnx::Conv_724)
%onnx::Conv_727 = Identity(%onnx::Conv_724)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%onnx::Conv_715 = Identity(%onnx::Conv_697)
%onnx::Conv_712 = Identity(%onnx::Conv_697)
%onnx::Conv_709 = Identity(%onnx::Conv_697)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_697)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%onnx::Conv_694 = Identity(%onnx::Conv_691)
%onnx::Conv_688 = Identity(%onnx::Conv_661)
%onnx::Conv_685 = Identity(%onnx::Conv_682)
%onnx::Conv_679 = Identity(%onnx::Conv_661)
%onnx::Conv_676 = Identity(%onnx::Conv_661)
%onnx::Conv_673 = Identity(%onnx::Conv_661)
%onnx::Conv_670 = Identity(%onnx::Conv_661)
%onnx::Conv_667 = Identity(%onnx::Conv_661)
%onnx::Conv_664 = Identity(%onnx::Conv_661)
%onnx::Conv_658 = Identity(%onnx::Conv_655)
%onnx::Conv_652 = Identity(%onnx::Conv_643)
%onnx::Conv_649 = Identity(%onnx::Conv_643)
%onnx::Conv_646 = Identity(%onnx::Conv_643)
%onnx::Conv_640 = Identity(%onnx::Conv_607)
%onnx::Conv_637 = Identity(%onnx::Conv_607)
%onnx::Conv_634 = Identity(%onnx::Conv_607)
%onnx::Conv_631 = Identity(%onnx::Conv_628)
%onnx::Conv_625 = Identity(%onnx::Conv_607)
%onnx::Conv_622 = Identity(%onnx::Conv_607)
%onnx::Conv_619 = Identity(%onnx::Conv_607)
%onnx::Conv_616 = Identity(%onnx::Conv_607)
%onnx::Conv_613 = Identity(%onnx::Conv_610)
%onnx::Conv_604 = Identity(%onnx::Conv_601)
%onnx::Conv_598 = Identity(%onnx::Conv_589)
%onnx::Conv_595 = Identity(%onnx::Conv_589)
%onnx::Conv_592 = Identity(%onnx::Conv_589)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_588, %onnx::Conv_589)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_591, %onnx::Conv_592)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_594, %onnx::Conv_595)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_597, %onnx::Conv_598)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_600, %onnx::Conv_601)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_603, %onnx::Conv_604)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_606, %onnx::Conv_607)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_609, %onnx::Conv_610)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_612, %onnx::Conv_613)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_615, %onnx::Conv_616)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%586 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %586
}
|
val_accuracy
| 0 | 60,242,304 | 1,670,452 |
{'zcp_synflow': 66.9789573823428, 'zcp_zen': 57.293643951416016, 'zcp_epe_nas': 6.867075209083241, 'zcp_fisher': 0.03958449885249138, 'zcp_flops': 60242304.0, 'zcp_grad_norm': 16.769073486328125, 'zcp_grasp': -0.034832000732421875, 'zcp_jacov': -16.046090133486402, 'zcp_l2_norm': 511.662109375, 'zcp_nwot': 211.12135796248538, 'zcp_params': 1670452.0, 'zcp_plain': -0.007717879489064217, 'zcp_snip': 27.041521072387695, 'lat_1080ti_1': 0.3693611871332916, 'lat_1080ti_32': 0.37379970028645015, 'lat_1080ti_64': 0.3870642202878051, 'lat_2080ti_1': 0.3598876949222621, 'lat_2080ti_32': 0.3966027583177001, 'lat_2080ti_64': 0.3734394502721573, 'lat_essential_ph_1': 0.22641509433962265, 'lat_eyeriss': 0.3622210404709075, 'lat_fpga': 0.2995181032861991, 'lat_gold_6226': 0.2550249177874168, 'lat_gold_6240': 0.3222149087423295, 'lat_pixel2': 0.21739130434782608, 'lat_pixel3': 0.40112908743071185, 'lat_raspi4': 0.41261373862731066, 'lat_samsung_a50': 0.1368421052631579, 'lat_samsung_s7': 0.11811023622047244, 'lat_silver_4114': 0.34365913814931737, 'lat_silver_4210r': 0.33667569703455247, 'lat_titan_rtx_1': 0.3444466493481029, 'lat_titan_rtx_32': 0.36404412596821767, 'lat_titan_rtx_64': 0.3691904677373943, 'lat_titanx_1': 0.1777383383697353, 'lat_titanx_32': 0.3719702108000963, 'lat_titanx_64': 0.3628581162352269, 'lat_titanxp_1': 0.3232864816103362, 'lat_titanxp_32': 0.37938611538464434, 'lat_titanxp_64': 0.3910943342651759}
| |
FBNet_3787
|
FBNet
|
3787
|
3787
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_614[FLOAT, 16x3x3x3]
%onnx::Conv_615[FLOAT, 16]
%onnx::Conv_617[FLOAT, 16x16x1x1]
%onnx::Conv_620[FLOAT, 16x1x5x5]
%onnx::Conv_623[FLOAT, 24x16x1x1]
%onnx::Conv_624[FLOAT, 24]
%onnx::Conv_626[FLOAT, 24x24x1x1]
%onnx::Conv_629[FLOAT, 24x1x5x5]
%onnx::Conv_632[FLOAT, 24x24x1x1]
%onnx::Conv_635[FLOAT, 24x12x1x1]
%onnx::Conv_638[FLOAT, 24x1x3x3]
%onnx::Conv_641[FLOAT, 24x12x1x1]
%onnx::Conv_644[FLOAT, 24x12x1x1]
%onnx::Conv_647[FLOAT, 24x1x5x5]
%onnx::Conv_650[FLOAT, 24x12x1x1]
%onnx::Conv_653[FLOAT, 72x24x1x1]
%onnx::Conv_654[FLOAT, 72]
%onnx::Conv_656[FLOAT, 72x1x3x3]
%onnx::Conv_659[FLOAT, 32x72x1x1]
%onnx::Conv_660[FLOAT, 32]
%onnx::Conv_662[FLOAT, 32x16x1x1]
%onnx::Conv_665[FLOAT, 32x1x5x5]
%onnx::Conv_668[FLOAT, 32x16x1x1]
%onnx::Conv_671[FLOAT, 32x16x1x1]
%onnx::Conv_674[FLOAT, 32x1x3x3]
%onnx::Conv_677[FLOAT, 32x16x1x1]
%onnx::Conv_680[FLOAT, 192x32x1x1]
%onnx::Conv_681[FLOAT, 192]
%onnx::Conv_683[FLOAT, 192x1x5x5]
%onnx::Conv_686[FLOAT, 32x192x1x1]
%onnx::Conv_689[FLOAT, 96x32x1x1]
%onnx::Conv_690[FLOAT, 96]
%onnx::Conv_692[FLOAT, 96x1x3x3]
%onnx::Conv_695[FLOAT, 64x96x1x1]
%onnx::Conv_696[FLOAT, 64]
%onnx::Conv_698[FLOAT, 384x64x1x1]
%onnx::Conv_699[FLOAT, 384]
%onnx::Conv_701[FLOAT, 384x1x5x5]
%onnx::Conv_704[FLOAT, 64x384x1x1]
%onnx::Conv_707[FLOAT, 64x32x1x1]
%onnx::Conv_710[FLOAT, 64x1x5x5]
%onnx::Conv_713[FLOAT, 64x32x1x1]
%onnx::Conv_716[FLOAT, 192x64x1x1]
%onnx::Conv_719[FLOAT, 192x1x5x5]
%onnx::Conv_722[FLOAT, 64x192x1x1]
%onnx::Conv_725[FLOAT, 192x64x1x1]
%onnx::Conv_728[FLOAT, 192x1x5x5]
%onnx::Conv_731[FLOAT, 112x192x1x1]
%onnx::Conv_732[FLOAT, 112]
%onnx::Conv_734[FLOAT, 672x112x1x1]
%onnx::Conv_735[FLOAT, 672]
%onnx::Conv_737[FLOAT, 672x1x5x5]
%onnx::Conv_740[FLOAT, 112x672x1x1]
%onnx::Conv_743[FLOAT, 112x112x1x1]
%onnx::Conv_746[FLOAT, 112x1x5x5]
%onnx::Conv_749[FLOAT, 112x112x1x1]
%onnx::Conv_752[FLOAT, 336x112x1x1]
%onnx::Conv_753[FLOAT, 336]
%onnx::Conv_755[FLOAT, 336x1x5x5]
%onnx::Conv_758[FLOAT, 184x336x1x1]
%onnx::Conv_759[FLOAT, 184]
%onnx::Conv_761[FLOAT, 184x184x1x1]
%onnx::Conv_764[FLOAT, 184x1x5x5]
%onnx::Conv_767[FLOAT, 184x184x1x1]
%onnx::Conv_770[FLOAT, 1104x184x1x1]
%onnx::Conv_771[FLOAT, 1104]
%onnx::Conv_773[FLOAT, 1104x1x3x3]
%onnx::Conv_776[FLOAT, 184x1104x1x1]
%onnx::Conv_779[FLOAT, 352x184x1x1]
%onnx::Conv_780[FLOAT, 352]
%onnx::Conv_782[FLOAT, 1504x352x1x1]
%onnx::Conv_783[FLOAT, 1504]
) {
%onnx::Conv_777 = Identity(%onnx::Conv_759)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_768 = Identity(%onnx::Conv_759)
%onnx::Conv_765 = Identity(%onnx::Conv_759)
%onnx::Conv_762 = Identity(%onnx::Conv_759)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_750 = Identity(%onnx::Conv_732)
%onnx::Conv_747 = Identity(%onnx::Conv_732)
%onnx::Conv_744 = Identity(%onnx::Conv_732)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_729 = Identity(%onnx::Conv_681)
%onnx::Conv_726 = Identity(%onnx::Conv_681)
%onnx::Conv_723 = Identity(%onnx::Conv_696)
%onnx::Conv_720 = Identity(%onnx::Conv_681)
%onnx::Conv_717 = Identity(%onnx::Conv_681)
%onnx::Conv_714 = Identity(%onnx::Conv_696)
%onnx::Conv_711 = Identity(%onnx::Conv_696)
%onnx::Conv_708 = Identity(%onnx::Conv_696)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%onnx::Conv_693 = Identity(%onnx::Conv_690)
%onnx::Conv_687 = Identity(%onnx::Conv_660)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_678 = Identity(%onnx::Conv_660)
%onnx::Conv_675 = Identity(%onnx::Conv_660)
%onnx::Conv_672 = Identity(%onnx::Conv_660)
%onnx::Conv_669 = Identity(%onnx::Conv_660)
%onnx::Conv_666 = Identity(%onnx::Conv_660)
%onnx::Conv_663 = Identity(%onnx::Conv_660)
%onnx::Conv_657 = Identity(%onnx::Conv_654)
%onnx::Conv_651 = Identity(%onnx::Conv_624)
%onnx::Conv_648 = Identity(%onnx::Conv_624)
%onnx::Conv_645 = Identity(%onnx::Conv_624)
%onnx::Conv_642 = Identity(%onnx::Conv_624)
%onnx::Conv_639 = Identity(%onnx::Conv_624)
%onnx::Conv_636 = Identity(%onnx::Conv_624)
%onnx::Conv_633 = Identity(%onnx::Conv_624)
%onnx::Conv_630 = Identity(%onnx::Conv_624)
%onnx::Conv_627 = Identity(%onnx::Conv_624)
%onnx::Conv_621 = Identity(%onnx::Conv_615)
%onnx::Conv_618 = Identity(%onnx::Conv_615)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_614, %onnx::Conv_615)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%612 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %612
}
|
val_accuracy
| 0 | 58,473,856 | 1,730,348 |
{'zcp_synflow': 72.60674506096167, 'zcp_zen': 61.930110931396484, 'zcp_epe_nas': 10.34997695320321, 'zcp_fisher': 0.04101071134209633, 'zcp_flops': 58473856.0, 'zcp_grad_norm': 16.92205238342285, 'zcp_grasp': -0.008307933807373047, 'zcp_jacov': -16.05364415596071, 'zcp_l2_norm': 553.1924438476562, 'zcp_nwot': 207.10991912240013, 'zcp_params': 1730348.0, 'zcp_plain': 9.423686424270272e-05, 'zcp_snip': 25.664487838745117, 'lat_1080ti_1': 0.3817778475556632, 'lat_1080ti_32': 0.32700103523681084, 'lat_1080ti_64': 0.24645160715673262, 'lat_2080ti_1': 0.4695580094167275, 'lat_2080ti_32': 0.3366241333110487, 'lat_2080ti_64': 0.2390619547805217, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.30198228595529314, 'lat_fpga': 0.3004185673715353, 'lat_gold_6226': 0.3150306995987031, 'lat_gold_6240': 0.43019471889636246, 'lat_pixel2': 0.34782608695652173, 'lat_pixel3': 0.3396930396014586, 'lat_raspi4': 0.34561561513734657, 'lat_samsung_a50': 0.1368421052631579, 'lat_samsung_s7': 0.11811023622047244, 'lat_silver_4114': 0.4473542332504433, 'lat_silver_4210r': 0.4606547120328451, 'lat_titan_rtx_1': 0.42821816672233615, 'lat_titan_rtx_32': 0.34112295627619055, 'lat_titan_rtx_64': 0.23662037561872926, 'lat_titanx_1': 0.22807333102467223, 'lat_titanx_32': 0.27389103142924937, 'lat_titanx_64': 0.24824174770066512, 'lat_titanxp_1': 0.40428528351393966, 'lat_titanxp_32': 0.2882685100103917, 'lat_titanxp_64': 0.22908002857689416}
| |
FBNet_4761
|
FBNet
|
4761
|
4761
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_652[FLOAT, 16x3x3x3]
%onnx::Conv_653[FLOAT, 16]
%onnx::Conv_655[FLOAT, 48x16x1x1]
%onnx::Conv_656[FLOAT, 48]
%onnx::Conv_658[FLOAT, 48x1x3x3]
%onnx::Conv_661[FLOAT, 16x48x1x1]
%onnx::Conv_664[FLOAT, 24x16x1x1]
%onnx::Conv_665[FLOAT, 24]
%onnx::Conv_667[FLOAT, 144x24x1x1]
%onnx::Conv_668[FLOAT, 144]
%onnx::Conv_670[FLOAT, 144x1x5x5]
%onnx::Conv_673[FLOAT, 24x144x1x1]
%onnx::Conv_676[FLOAT, 24x24x1x1]
%onnx::Conv_679[FLOAT, 24x1x5x5]
%onnx::Conv_682[FLOAT, 24x24x1x1]
%onnx::Conv_685[FLOAT, 72x24x1x1]
%onnx::Conv_686[FLOAT, 72]
%onnx::Conv_688[FLOAT, 72x1x5x5]
%onnx::Conv_691[FLOAT, 32x72x1x1]
%onnx::Conv_692[FLOAT, 32]
%onnx::Conv_694[FLOAT, 192x32x1x1]
%onnx::Conv_695[FLOAT, 192]
%onnx::Conv_697[FLOAT, 192x1x3x3]
%onnx::Conv_700[FLOAT, 32x192x1x1]
%onnx::Conv_703[FLOAT, 32x16x1x1]
%onnx::Conv_706[FLOAT, 32x1x3x3]
%onnx::Conv_709[FLOAT, 32x16x1x1]
%onnx::Conv_712[FLOAT, 32x32x1x1]
%onnx::Conv_715[FLOAT, 32x1x3x3]
%onnx::Conv_718[FLOAT, 32x32x1x1]
%onnx::Conv_721[FLOAT, 96x32x1x1]
%onnx::Conv_722[FLOAT, 96]
%onnx::Conv_724[FLOAT, 96x1x5x5]
%onnx::Conv_727[FLOAT, 64x96x1x1]
%onnx::Conv_728[FLOAT, 64]
%onnx::Conv_730[FLOAT, 192x64x1x1]
%onnx::Conv_733[FLOAT, 192x1x3x3]
%onnx::Conv_736[FLOAT, 64x192x1x1]
%onnx::Conv_739[FLOAT, 64x64x1x1]
%onnx::Conv_742[FLOAT, 64x1x5x5]
%onnx::Conv_745[FLOAT, 64x64x1x1]
%onnx::Conv_748[FLOAT, 192x64x1x1]
%onnx::Conv_751[FLOAT, 192x1x3x3]
%onnx::Conv_754[FLOAT, 64x192x1x1]
%onnx::Conv_757[FLOAT, 64x32x1x1]
%onnx::Conv_760[FLOAT, 64x1x3x3]
%onnx::Conv_763[FLOAT, 112x32x1x1]
%onnx::Conv_764[FLOAT, 112]
%onnx::Conv_766[FLOAT, 336x112x1x1]
%onnx::Conv_767[FLOAT, 336]
%onnx::Conv_769[FLOAT, 336x1x3x3]
%onnx::Conv_772[FLOAT, 112x336x1x1]
%onnx::Conv_775[FLOAT, 112x56x1x1]
%onnx::Conv_778[FLOAT, 112x1x5x5]
%onnx::Conv_781[FLOAT, 112x56x1x1]
%onnx::Conv_784[FLOAT, 112x56x1x1]
%onnx::Conv_787[FLOAT, 112x1x3x3]
%onnx::Conv_790[FLOAT, 112x56x1x1]
%onnx::Conv_793[FLOAT, 184x112x1x1]
%onnx::Conv_794[FLOAT, 184]
%onnx::Conv_796[FLOAT, 1104x184x1x1]
%onnx::Conv_797[FLOAT, 1104]
%onnx::Conv_799[FLOAT, 1104x1x5x5]
%onnx::Conv_802[FLOAT, 184x1104x1x1]
%onnx::Conv_805[FLOAT, 552x184x1x1]
%onnx::Conv_806[FLOAT, 552]
%onnx::Conv_808[FLOAT, 552x1x5x5]
%onnx::Conv_811[FLOAT, 184x552x1x1]
%onnx::Conv_814[FLOAT, 184x92x1x1]
%onnx::Conv_817[FLOAT, 184x1x3x3]
%onnx::Conv_820[FLOAT, 184x92x1x1]
%onnx::Conv_823[FLOAT, 1104x184x1x1]
%onnx::Conv_826[FLOAT, 1104x1x5x5]
%onnx::Conv_829[FLOAT, 352x1104x1x1]
%onnx::Conv_830[FLOAT, 352]
%onnx::Conv_832[FLOAT, 1504x352x1x1]
%onnx::Conv_833[FLOAT, 1504]
) {
%onnx::Conv_827 = Identity(%onnx::Conv_797)
%onnx::Conv_824 = Identity(%onnx::Conv_797)
%onnx::Conv_821 = Identity(%onnx::Conv_794)
%onnx::Conv_818 = Identity(%onnx::Conv_794)
%onnx::Conv_815 = Identity(%onnx::Conv_794)
%onnx::Conv_812 = Identity(%onnx::Conv_794)
%onnx::Conv_809 = Identity(%onnx::Conv_806)
%onnx::Conv_803 = Identity(%onnx::Conv_794)
%onnx::Conv_800 = Identity(%onnx::Conv_797)
%onnx::Conv_791 = Identity(%onnx::Conv_764)
%onnx::Conv_788 = Identity(%onnx::Conv_764)
%onnx::Conv_785 = Identity(%onnx::Conv_764)
%onnx::Conv_782 = Identity(%onnx::Conv_764)
%onnx::Conv_779 = Identity(%onnx::Conv_764)
%onnx::Conv_776 = Identity(%onnx::Conv_764)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_767)
%onnx::Conv_761 = Identity(%onnx::Conv_728)
%onnx::Conv_758 = Identity(%onnx::Conv_728)
%onnx::Conv_755 = Identity(%onnx::Conv_728)
%onnx::Conv_752 = Identity(%onnx::Conv_695)
%onnx::Conv_749 = Identity(%onnx::Conv_695)
%onnx::Conv_746 = Identity(%onnx::Conv_728)
%onnx::Conv_743 = Identity(%onnx::Conv_728)
%onnx::Conv_740 = Identity(%onnx::Conv_728)
%onnx::Conv_737 = Identity(%onnx::Conv_728)
%onnx::Conv_734 = Identity(%onnx::Conv_695)
%onnx::Conv_731 = Identity(%onnx::Conv_695)
%onnx::Conv_725 = Identity(%onnx::Conv_722)
%onnx::Conv_719 = Identity(%onnx::Conv_692)
%onnx::Conv_716 = Identity(%onnx::Conv_692)
%onnx::Conv_713 = Identity(%onnx::Conv_692)
%onnx::Conv_710 = Identity(%onnx::Conv_692)
%onnx::Conv_707 = Identity(%onnx::Conv_692)
%onnx::Conv_704 = Identity(%onnx::Conv_692)
%onnx::Conv_701 = Identity(%onnx::Conv_692)
%onnx::Conv_698 = Identity(%onnx::Conv_695)
%onnx::Conv_689 = Identity(%onnx::Conv_686)
%onnx::Conv_683 = Identity(%onnx::Conv_665)
%onnx::Conv_680 = Identity(%onnx::Conv_665)
%onnx::Conv_677 = Identity(%onnx::Conv_665)
%onnx::Conv_674 = Identity(%onnx::Conv_665)
%onnx::Conv_671 = Identity(%onnx::Conv_668)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_656)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_652, %onnx::Conv_653)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.1/relu/Relu_output_0 = Relu(%/cells.1/conv/Conv_output_0)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/relu/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/relu/Relu_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%650 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %650
}
|
val_accuracy
| 0 | 68,002,560 | 2,257,812 |
{'zcp_synflow': 76.02036291083549, 'zcp_zen': 66.52484130859375, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.16345638036727905, 'zcp_flops': 68002560.0, 'zcp_grad_norm': 24.069957733154297, 'zcp_grasp': 0.06146240234375, 'zcp_jacov': -16.058298996684943, 'zcp_l2_norm': 620.9983520507812, 'zcp_nwot': 211.3472710157123, 'zcp_params': 2257812.0, 'zcp_plain': -0.0022134180180728436, 'zcp_snip': 45.321353912353516, 'lat_1080ti_1': 0.6566524211077261, 'lat_1080ti_32': 0.5357066428298203, 'lat_1080ti_64': 0.4648606662653841, 'lat_2080ti_1': 0.6128165706989621, 'lat_2080ti_32': 0.5908632734547195, 'lat_2080ti_64': 0.47859517520950234, 'lat_essential_ph_1': 0.32075471698113206, 'lat_eyeriss': 0.4571543835842517, 'lat_fpga': 0.46025589012359114, 'lat_gold_6226': 0.4101457020789089, 'lat_gold_6240': 0.4984698219543191, 'lat_pixel2': 0.30434782608695654, 'lat_pixel3': 0.4501759997956509, 'lat_raspi4': 0.5412326096843686, 'lat_samsung_a50': 0.2, 'lat_samsung_s7': 0.18110236220472442, 'lat_silver_4114': 0.5061335537949256, 'lat_silver_4210r': 0.5148192377393034, 'lat_titan_rtx_1': 0.5739592941518602, 'lat_titan_rtx_32': 0.5795397642173541, 'lat_titan_rtx_64': 0.47860473840950274, 'lat_titanx_1': 0.31290792755575797, 'lat_titanx_32': 0.5210418218638854, 'lat_titanx_64': 0.42976168964795625, 'lat_titanxp_1': 0.5601474655880583, 'lat_titanxp_32': 0.5371026784106258, 'lat_titanxp_64': 0.460784222181696}
| |
FBNet_2932
|
FBNet
|
2932
|
2932
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_640[FLOAT, 16x3x3x3]
%onnx::Conv_641[FLOAT, 16]
%onnx::Conv_643[FLOAT, 48x16x1x1]
%onnx::Conv_644[FLOAT, 48]
%onnx::Conv_646[FLOAT, 48x1x5x5]
%onnx::Conv_649[FLOAT, 16x48x1x1]
%onnx::Conv_652[FLOAT, 16x8x1x1]
%onnx::Conv_655[FLOAT, 16x1x5x5]
%onnx::Conv_658[FLOAT, 24x8x1x1]
%onnx::Conv_659[FLOAT, 24]
%onnx::Conv_661[FLOAT, 72x24x1x1]
%onnx::Conv_662[FLOAT, 72]
%onnx::Conv_664[FLOAT, 72x1x5x5]
%onnx::Conv_667[FLOAT, 24x72x1x1]
%onnx::Conv_670[FLOAT, 72x24x1x1]
%onnx::Conv_673[FLOAT, 72x1x3x3]
%onnx::Conv_676[FLOAT, 24x72x1x1]
%onnx::Conv_679[FLOAT, 24x24x1x1]
%onnx::Conv_682[FLOAT, 24x1x5x5]
%onnx::Conv_685[FLOAT, 24x24x1x1]
%onnx::Conv_688[FLOAT, 72x24x1x1]
%onnx::Conv_691[FLOAT, 72x1x3x3]
%onnx::Conv_694[FLOAT, 32x72x1x1]
%onnx::Conv_695[FLOAT, 32]
%onnx::Conv_697[FLOAT, 32x16x1x1]
%onnx::Conv_700[FLOAT, 32x1x3x3]
%onnx::Conv_703[FLOAT, 32x16x1x1]
%onnx::Conv_706[FLOAT, 192x32x1x1]
%onnx::Conv_707[FLOAT, 192]
%onnx::Conv_709[FLOAT, 192x1x5x5]
%onnx::Conv_712[FLOAT, 32x192x1x1]
%onnx::Conv_715[FLOAT, 32x16x1x1]
%onnx::Conv_718[FLOAT, 32x1x5x5]
%onnx::Conv_721[FLOAT, 64x16x1x1]
%onnx::Conv_722[FLOAT, 64]
%onnx::Conv_724[FLOAT, 192x64x1x1]
%onnx::Conv_727[FLOAT, 192x1x5x5]
%onnx::Conv_730[FLOAT, 64x192x1x1]
%onnx::Conv_733[FLOAT, 384x64x1x1]
%onnx::Conv_734[FLOAT, 384]
%onnx::Conv_736[FLOAT, 384x1x5x5]
%onnx::Conv_739[FLOAT, 64x384x1x1]
%onnx::Conv_742[FLOAT, 192x64x1x1]
%onnx::Conv_745[FLOAT, 192x1x5x5]
%onnx::Conv_748[FLOAT, 112x192x1x1]
%onnx::Conv_749[FLOAT, 112]
%onnx::Conv_751[FLOAT, 112x112x1x1]
%onnx::Conv_754[FLOAT, 112x1x3x3]
%onnx::Conv_757[FLOAT, 112x112x1x1]
%onnx::Conv_760[FLOAT, 112x112x1x1]
%onnx::Conv_763[FLOAT, 112x1x5x5]
%onnx::Conv_766[FLOAT, 112x112x1x1]
%onnx::Conv_769[FLOAT, 336x112x1x1]
%onnx::Conv_770[FLOAT, 336]
%onnx::Conv_772[FLOAT, 336x1x5x5]
%onnx::Conv_775[FLOAT, 112x336x1x1]
%onnx::Conv_778[FLOAT, 112x56x1x1]
%onnx::Conv_781[FLOAT, 112x1x5x5]
%onnx::Conv_784[FLOAT, 184x56x1x1]
%onnx::Conv_785[FLOAT, 184]
%onnx::Conv_787[FLOAT, 184x184x1x1]
%onnx::Conv_790[FLOAT, 184x1x5x5]
%onnx::Conv_793[FLOAT, 184x184x1x1]
%onnx::Conv_796[FLOAT, 1104x184x1x1]
%onnx::Conv_797[FLOAT, 1104]
%onnx::Conv_799[FLOAT, 1104x1x3x3]
%onnx::Conv_802[FLOAT, 184x1104x1x1]
%onnx::Conv_805[FLOAT, 1104x184x1x1]
%onnx::Conv_808[FLOAT, 1104x1x3x3]
%onnx::Conv_811[FLOAT, 184x1104x1x1]
%onnx::Conv_814[FLOAT, 552x184x1x1]
%onnx::Conv_815[FLOAT, 552]
%onnx::Conv_817[FLOAT, 552x1x5x5]
%onnx::Conv_820[FLOAT, 352x552x1x1]
%onnx::Conv_821[FLOAT, 352]
%onnx::Conv_823[FLOAT, 1504x352x1x1]
%onnx::Conv_824[FLOAT, 1504]
) {
%onnx::Conv_818 = Identity(%onnx::Conv_815)
%onnx::Conv_812 = Identity(%onnx::Conv_785)
%onnx::Conv_809 = Identity(%onnx::Conv_797)
%onnx::Conv_806 = Identity(%onnx::Conv_797)
%onnx::Conv_803 = Identity(%onnx::Conv_785)
%onnx::Conv_800 = Identity(%onnx::Conv_797)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_785)
%onnx::Conv_788 = Identity(%onnx::Conv_785)
%onnx::Conv_782 = Identity(%onnx::Conv_749)
%onnx::Conv_779 = Identity(%onnx::Conv_749)
%onnx::Conv_776 = Identity(%onnx::Conv_749)
%onnx::Conv_773 = Identity(%onnx::Conv_770)
%onnx::Conv_767 = Identity(%onnx::Conv_749)
%onnx::Conv_764 = Identity(%onnx::Conv_749)
%onnx::Conv_761 = Identity(%onnx::Conv_749)
%onnx::Conv_758 = Identity(%onnx::Conv_749)
%onnx::Conv_755 = Identity(%onnx::Conv_749)
%onnx::Conv_752 = Identity(%onnx::Conv_749)
%onnx::Conv_746 = Identity(%onnx::Conv_707)
%onnx::Conv_743 = Identity(%onnx::Conv_707)
%onnx::Conv_740 = Identity(%onnx::Conv_722)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_731 = Identity(%onnx::Conv_722)
%onnx::Conv_728 = Identity(%onnx::Conv_707)
%onnx::Conv_725 = Identity(%onnx::Conv_707)
%onnx::Conv_719 = Identity(%onnx::Conv_695)
%onnx::Conv_716 = Identity(%onnx::Conv_695)
%onnx::Conv_713 = Identity(%onnx::Conv_695)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_695)
%onnx::Conv_701 = Identity(%onnx::Conv_695)
%onnx::Conv_698 = Identity(%onnx::Conv_695)
%onnx::Conv_692 = Identity(%onnx::Conv_662)
%onnx::Conv_689 = Identity(%onnx::Conv_662)
%onnx::Conv_686 = Identity(%onnx::Conv_659)
%onnx::Conv_683 = Identity(%onnx::Conv_659)
%onnx::Conv_680 = Identity(%onnx::Conv_659)
%onnx::Conv_677 = Identity(%onnx::Conv_659)
%onnx::Conv_674 = Identity(%onnx::Conv_662)
%onnx::Conv_671 = Identity(%onnx::Conv_662)
%onnx::Conv_668 = Identity(%onnx::Conv_659)
%onnx::Conv_665 = Identity(%onnx::Conv_662)
%onnx::Conv_656 = Identity(%onnx::Conv_641)
%onnx::Conv_653 = Identity(%onnx::Conv_641)
%onnx::Conv_650 = Identity(%onnx::Conv_641)
%onnx::Conv_647 = Identity(%onnx::Conv_644)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_640, %onnx::Conv_641)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%638 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %638
}
|
val_accuracy
| 0 | 71,346,048 | 2,245,156 |
{'zcp_synflow': 78.59587537525937, 'zcp_zen': 69.66002655029297, 'zcp_epe_nas': 23.761173600120927, 'zcp_fisher': 0.09556645154953003, 'zcp_flops': 71346048.0, 'zcp_grad_norm': 25.343204498291016, 'zcp_grasp': 0.034770965576171875, 'zcp_jacov': -16.052355329025218, 'zcp_l2_norm': 648.0650634765625, 'zcp_nwot': 211.61969447717476, 'zcp_params': 2245156.0, 'zcp_plain': -0.005977467633783817, 'zcp_snip': 43.257057189941406, 'lat_1080ti_1': 0.56098198288377, 'lat_1080ti_32': 0.4831678448190079, 'lat_1080ti_64': 0.4491121416604629, 'lat_2080ti_1': 0.607423624904659, 'lat_2080ti_32': 0.4838250848702368, 'lat_2080ti_64': 0.4256831089311464, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.49964241835219037, 'lat_fpga': 0.5204979269535616, 'lat_gold_6226': 0.4332820356968391, 'lat_gold_6240': 0.5654683373325865, 'lat_pixel2': 0.32608695652173914, 'lat_pixel3': 0.4945745407037814, 'lat_raspi4': 0.5660440694675638, 'lat_samsung_a50': 0.22105263157894736, 'lat_samsung_s7': 0.1968503937007874, 'lat_silver_4114': 0.5978664704152291, 'lat_silver_4210r': 0.5767715007098205, 'lat_titan_rtx_1': 0.5655918091246986, 'lat_titan_rtx_32': 0.47691287540390176, 'lat_titan_rtx_64': 0.435701604278797, 'lat_titanx_1': 0.31243475893768474, 'lat_titanx_32': 0.45529713415558476, 'lat_titanx_64': 0.4456304399665406, 'lat_titanxp_1': 0.5354417331571701, 'lat_titanxp_32': 0.46695854160321265, 'lat_titanxp_64': 0.43705918988346115}
| |
FBNet_4756
|
FBNet
|
4756
|
4756
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_724[FLOAT, 16x3x3x3]
%onnx::Conv_725[FLOAT, 16]
%onnx::Conv_727[FLOAT, 16x16x1x1]
%onnx::Conv_730[FLOAT, 16x1x5x5]
%onnx::Conv_733[FLOAT, 16x16x1x1]
%onnx::Conv_736[FLOAT, 48x16x1x1]
%onnx::Conv_737[FLOAT, 48]
%onnx::Conv_739[FLOAT, 48x1x5x5]
%onnx::Conv_742[FLOAT, 24x48x1x1]
%onnx::Conv_743[FLOAT, 24]
%onnx::Conv_745[FLOAT, 24x12x1x1]
%onnx::Conv_748[FLOAT, 24x1x3x3]
%onnx::Conv_751[FLOAT, 24x12x1x1]
%onnx::Conv_754[FLOAT, 72x24x1x1]
%onnx::Conv_755[FLOAT, 72]
%onnx::Conv_757[FLOAT, 72x1x3x3]
%onnx::Conv_760[FLOAT, 24x72x1x1]
%onnx::Conv_763[FLOAT, 144x24x1x1]
%onnx::Conv_764[FLOAT, 144]
%onnx::Conv_766[FLOAT, 144x1x5x5]
%onnx::Conv_769[FLOAT, 24x144x1x1]
%onnx::Conv_772[FLOAT, 32x24x1x1]
%onnx::Conv_773[FLOAT, 32]
%onnx::Conv_775[FLOAT, 96x32x1x1]
%onnx::Conv_776[FLOAT, 96]
%onnx::Conv_778[FLOAT, 96x1x5x5]
%onnx::Conv_781[FLOAT, 32x96x1x1]
%onnx::Conv_784[FLOAT, 96x32x1x1]
%onnx::Conv_787[FLOAT, 96x1x3x3]
%onnx::Conv_790[FLOAT, 32x96x1x1]
%onnx::Conv_793[FLOAT, 32x16x1x1]
%onnx::Conv_796[FLOAT, 32x1x5x5]
%onnx::Conv_799[FLOAT, 64x16x1x1]
%onnx::Conv_800[FLOAT, 64]
%onnx::Conv_802[FLOAT, 192x64x1x1]
%onnx::Conv_803[FLOAT, 192]
%onnx::Conv_805[FLOAT, 192x1x3x3]
%onnx::Conv_808[FLOAT, 64x192x1x1]
%onnx::Conv_811[FLOAT, 64x32x1x1]
%onnx::Conv_814[FLOAT, 64x1x3x3]
%onnx::Conv_817[FLOAT, 64x32x1x1]
%onnx::Conv_820[FLOAT, 64x32x1x1]
%onnx::Conv_823[FLOAT, 64x1x3x3]
%onnx::Conv_826[FLOAT, 64x32x1x1]
%onnx::Conv_829[FLOAT, 64x64x1x1]
%onnx::Conv_832[FLOAT, 64x1x5x5]
%onnx::Conv_835[FLOAT, 112x64x1x1]
%onnx::Conv_836[FLOAT, 112]
%onnx::Conv_838[FLOAT, 112x112x1x1]
%onnx::Conv_841[FLOAT, 112x1x5x5]
%onnx::Conv_844[FLOAT, 112x112x1x1]
%onnx::Conv_847[FLOAT, 112x112x1x1]
%onnx::Conv_850[FLOAT, 112x1x3x3]
%onnx::Conv_853[FLOAT, 112x112x1x1]
%onnx::Conv_856[FLOAT, 112x56x1x1]
%onnx::Conv_859[FLOAT, 112x1x5x5]
%onnx::Conv_862[FLOAT, 112x56x1x1]
%onnx::Conv_865[FLOAT, 112x56x1x1]
%onnx::Conv_868[FLOAT, 112x1x3x3]
%onnx::Conv_871[FLOAT, 184x56x1x1]
%onnx::Conv_872[FLOAT, 184]
%onnx::Conv_874[FLOAT, 552x184x1x1]
%onnx::Conv_875[FLOAT, 552]
%onnx::Conv_877[FLOAT, 552x1x5x5]
%onnx::Conv_880[FLOAT, 184x552x1x1]
%onnx::Conv_883[FLOAT, 1104x184x1x1]
%onnx::Conv_884[FLOAT, 1104]
%onnx::Conv_886[FLOAT, 1104x1x3x3]
%onnx::Conv_889[FLOAT, 184x1104x1x1]
%onnx::Conv_892[FLOAT, 184x92x1x1]
%onnx::Conv_895[FLOAT, 184x1x5x5]
%onnx::Conv_898[FLOAT, 184x92x1x1]
%onnx::Conv_901[FLOAT, 184x92x1x1]
%onnx::Conv_904[FLOAT, 184x1x3x3]
%onnx::Conv_907[FLOAT, 352x92x1x1]
%onnx::Conv_908[FLOAT, 352]
%onnx::Conv_910[FLOAT, 1504x352x1x1]
%onnx::Conv_911[FLOAT, 1504]
) {
%onnx::Conv_905 = Identity(%onnx::Conv_872)
%onnx::Conv_902 = Identity(%onnx::Conv_872)
%onnx::Conv_899 = Identity(%onnx::Conv_872)
%onnx::Conv_896 = Identity(%onnx::Conv_872)
%onnx::Conv_893 = Identity(%onnx::Conv_872)
%onnx::Conv_890 = Identity(%onnx::Conv_872)
%onnx::Conv_887 = Identity(%onnx::Conv_884)
%onnx::Conv_881 = Identity(%onnx::Conv_872)
%onnx::Conv_878 = Identity(%onnx::Conv_875)
%onnx::Conv_869 = Identity(%onnx::Conv_836)
%onnx::Conv_866 = Identity(%onnx::Conv_836)
%onnx::Conv_863 = Identity(%onnx::Conv_836)
%onnx::Conv_860 = Identity(%onnx::Conv_836)
%onnx::Conv_857 = Identity(%onnx::Conv_836)
%onnx::Conv_854 = Identity(%onnx::Conv_836)
%onnx::Conv_851 = Identity(%onnx::Conv_836)
%onnx::Conv_848 = Identity(%onnx::Conv_836)
%onnx::Conv_845 = Identity(%onnx::Conv_836)
%onnx::Conv_842 = Identity(%onnx::Conv_836)
%onnx::Conv_839 = Identity(%onnx::Conv_836)
%onnx::Conv_833 = Identity(%onnx::Conv_800)
%onnx::Conv_830 = Identity(%onnx::Conv_800)
%onnx::Conv_827 = Identity(%onnx::Conv_800)
%onnx::Conv_824 = Identity(%onnx::Conv_800)
%onnx::Conv_821 = Identity(%onnx::Conv_800)
%onnx::Conv_818 = Identity(%onnx::Conv_800)
%onnx::Conv_815 = Identity(%onnx::Conv_800)
%onnx::Conv_812 = Identity(%onnx::Conv_800)
%onnx::Conv_809 = Identity(%onnx::Conv_800)
%onnx::Conv_806 = Identity(%onnx::Conv_803)
%onnx::Conv_797 = Identity(%onnx::Conv_773)
%onnx::Conv_794 = Identity(%onnx::Conv_773)
%onnx::Conv_791 = Identity(%onnx::Conv_773)
%onnx::Conv_788 = Identity(%onnx::Conv_776)
%onnx::Conv_785 = Identity(%onnx::Conv_776)
%onnx::Conv_782 = Identity(%onnx::Conv_773)
%onnx::Conv_779 = Identity(%onnx::Conv_776)
%onnx::Conv_770 = Identity(%onnx::Conv_743)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_761 = Identity(%onnx::Conv_743)
%onnx::Conv_758 = Identity(%onnx::Conv_755)
%onnx::Conv_752 = Identity(%onnx::Conv_743)
%onnx::Conv_749 = Identity(%onnx::Conv_743)
%onnx::Conv_746 = Identity(%onnx::Conv_743)
%onnx::Conv_740 = Identity(%onnx::Conv_737)
%onnx::Conv_734 = Identity(%onnx::Conv_725)
%onnx::Conv_731 = Identity(%onnx::Conv_725)
%onnx::Conv_728 = Identity(%onnx::Conv_725)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_724, %onnx::Conv_725)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.4/Add_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_862, %onnx::Conv_863)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_865, %onnx::Conv_866)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_868, %onnx::Conv_869)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_871, %onnx::Conv_872)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_874, %onnx::Conv_875)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_877, %onnx::Conv_878)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_880, %onnx::Conv_881)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_883, %onnx::Conv_884)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_886, %onnx::Conv_887)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_889, %onnx::Conv_890)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_892, %onnx::Conv_893)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_895, %onnx::Conv_896)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_898, %onnx::Conv_899)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_901, %onnx::Conv_902)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_904, %onnx::Conv_905)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_907, %onnx::Conv_908)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_910, %onnx::Conv_911)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%722 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %722
}
|
val_accuracy
| 0 | 55,874,688 | 1,597,004 |
{'zcp_synflow': 74.03673868058465, 'zcp_zen': 64.94268035888672, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.11792801320552826, 'zcp_flops': 55874688.0, 'zcp_grad_norm': 23.186365127563477, 'zcp_grasp': -0.06903457641601562, 'zcp_jacov': -16.073096615504003, 'zcp_l2_norm': 561.0025024414062, 'zcp_nwot': 210.7096969573796, 'zcp_params': 1597004.0, 'zcp_plain': -0.0051051960326731205, 'zcp_snip': 37.34600830078125, 'lat_1080ti_1': 0.7334891748537586, 'lat_1080ti_32': 0.672722097932278, 'lat_1080ti_64': 0.5507400678680455, 'lat_2080ti_1': 0.7411907927414805, 'lat_2080ti_32': 0.6865346937722455, 'lat_2080ti_64': 0.5529232268855268, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.33946417765389775, 'lat_fpga': 0.30735510939154365, 'lat_gold_6226': 0.21031977409927421, 'lat_gold_6240': 0.37383515452982063, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.3560844580607924, 'lat_raspi4': 0.3826562793579981, 'lat_samsung_a50': 0.14736842105263157, 'lat_samsung_s7': 0.13385826771653545, 'lat_silver_4114': 0.4029982812255307, 'lat_silver_4210r': 0.4284386311891637, 'lat_titan_rtx_1': 0.6885925191077785, 'lat_titan_rtx_32': 0.654532486591286, 'lat_titan_rtx_64': 0.5957069993967555, 'lat_titanx_1': 0.36779365156432536, 'lat_titanx_32': 0.6192403312508098, 'lat_titanx_64': 0.4930989165008977, 'lat_titanxp_1': 0.659420834529534, 'lat_titanxp_32': 0.6685469996048445, 'lat_titanxp_64': 0.5515568407892681}
| |
FBNet_1311
|
FBNet
|
1311
|
1311
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_577[FLOAT, 16x3x3x3]
%onnx::Conv_578[FLOAT, 16]
%onnx::Conv_580[FLOAT, 16x16x1x1]
%onnx::Conv_583[FLOAT, 16x1x5x5]
%onnx::Conv_586[FLOAT, 16x16x1x1]
%onnx::Conv_589[FLOAT, 16x16x1x1]
%onnx::Conv_592[FLOAT, 16x1x3x3]
%onnx::Conv_595[FLOAT, 24x16x1x1]
%onnx::Conv_596[FLOAT, 24]
%onnx::Conv_598[FLOAT, 24x12x1x1]
%onnx::Conv_601[FLOAT, 24x1x5x5]
%onnx::Conv_604[FLOAT, 24x12x1x1]
%onnx::Conv_607[FLOAT, 24x12x1x1]
%onnx::Conv_610[FLOAT, 24x1x5x5]
%onnx::Conv_613[FLOAT, 24x12x1x1]
%onnx::Conv_616[FLOAT, 24x24x1x1]
%onnx::Conv_619[FLOAT, 24x1x5x5]
%onnx::Conv_622[FLOAT, 24x24x1x1]
%onnx::Conv_625[FLOAT, 144x24x1x1]
%onnx::Conv_626[FLOAT, 144]
%onnx::Conv_628[FLOAT, 144x1x3x3]
%onnx::Conv_631[FLOAT, 32x144x1x1]
%onnx::Conv_632[FLOAT, 32]
%onnx::Conv_634[FLOAT, 96x32x1x1]
%onnx::Conv_635[FLOAT, 96]
%onnx::Conv_637[FLOAT, 96x1x3x3]
%onnx::Conv_640[FLOAT, 32x96x1x1]
%onnx::Conv_643[FLOAT, 192x32x1x1]
%onnx::Conv_644[FLOAT, 192]
%onnx::Conv_646[FLOAT, 192x1x5x5]
%onnx::Conv_649[FLOAT, 32x192x1x1]
%onnx::Conv_652[FLOAT, 192x32x1x1]
%onnx::Conv_655[FLOAT, 192x1x5x5]
%onnx::Conv_658[FLOAT, 64x192x1x1]
%onnx::Conv_659[FLOAT, 64]
%onnx::Conv_661[FLOAT, 192x64x1x1]
%onnx::Conv_664[FLOAT, 192x1x5x5]
%onnx::Conv_667[FLOAT, 64x192x1x1]
%onnx::Conv_670[FLOAT, 64x32x1x1]
%onnx::Conv_673[FLOAT, 64x1x5x5]
%onnx::Conv_676[FLOAT, 64x32x1x1]
%onnx::Conv_679[FLOAT, 384x64x1x1]
%onnx::Conv_680[FLOAT, 384]
%onnx::Conv_682[FLOAT, 384x1x3x3]
%onnx::Conv_685[FLOAT, 64x384x1x1]
%onnx::Conv_688[FLOAT, 64x64x1x1]
%onnx::Conv_691[FLOAT, 64x1x5x5]
%onnx::Conv_694[FLOAT, 112x64x1x1]
%onnx::Conv_695[FLOAT, 112]
%onnx::Conv_697[FLOAT, 112x112x1x1]
%onnx::Conv_700[FLOAT, 112x1x5x5]
%onnx::Conv_703[FLOAT, 112x112x1x1]
%onnx::Conv_706[FLOAT, 336x112x1x1]
%onnx::Conv_707[FLOAT, 336]
%onnx::Conv_709[FLOAT, 336x1x3x3]
%onnx::Conv_712[FLOAT, 112x336x1x1]
%onnx::Conv_715[FLOAT, 336x112x1x1]
%onnx::Conv_718[FLOAT, 336x1x3x3]
%onnx::Conv_721[FLOAT, 184x336x1x1]
%onnx::Conv_722[FLOAT, 184]
%onnx::Conv_724[FLOAT, 1104x184x1x1]
%onnx::Conv_725[FLOAT, 1104]
%onnx::Conv_727[FLOAT, 1104x1x5x5]
%onnx::Conv_730[FLOAT, 184x1104x1x1]
%onnx::Conv_733[FLOAT, 184x184x1x1]
%onnx::Conv_736[FLOAT, 184x1x3x3]
%onnx::Conv_739[FLOAT, 184x184x1x1]
%onnx::Conv_742[FLOAT, 352x184x1x1]
%onnx::Conv_743[FLOAT, 352]
%onnx::Conv_745[FLOAT, 1504x352x1x1]
%onnx::Conv_746[FLOAT, 1504]
) {
%onnx::Conv_740 = Identity(%onnx::Conv_722)
%onnx::Conv_737 = Identity(%onnx::Conv_722)
%onnx::Conv_734 = Identity(%onnx::Conv_722)
%onnx::Conv_731 = Identity(%onnx::Conv_722)
%onnx::Conv_728 = Identity(%onnx::Conv_725)
%onnx::Conv_719 = Identity(%onnx::Conv_707)
%onnx::Conv_716 = Identity(%onnx::Conv_707)
%onnx::Conv_713 = Identity(%onnx::Conv_695)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_695)
%onnx::Conv_701 = Identity(%onnx::Conv_695)
%onnx::Conv_698 = Identity(%onnx::Conv_695)
%onnx::Conv_692 = Identity(%onnx::Conv_659)
%onnx::Conv_689 = Identity(%onnx::Conv_659)
%onnx::Conv_686 = Identity(%onnx::Conv_659)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_677 = Identity(%onnx::Conv_659)
%onnx::Conv_674 = Identity(%onnx::Conv_659)
%onnx::Conv_671 = Identity(%onnx::Conv_659)
%onnx::Conv_668 = Identity(%onnx::Conv_659)
%onnx::Conv_665 = Identity(%onnx::Conv_644)
%onnx::Conv_662 = Identity(%onnx::Conv_644)
%onnx::Conv_656 = Identity(%onnx::Conv_644)
%onnx::Conv_653 = Identity(%onnx::Conv_644)
%onnx::Conv_650 = Identity(%onnx::Conv_632)
%onnx::Conv_647 = Identity(%onnx::Conv_644)
%onnx::Conv_641 = Identity(%onnx::Conv_632)
%onnx::Conv_638 = Identity(%onnx::Conv_635)
%onnx::Conv_629 = Identity(%onnx::Conv_626)
%onnx::Conv_623 = Identity(%onnx::Conv_596)
%onnx::Conv_620 = Identity(%onnx::Conv_596)
%onnx::Conv_617 = Identity(%onnx::Conv_596)
%onnx::Conv_614 = Identity(%onnx::Conv_596)
%onnx::Conv_611 = Identity(%onnx::Conv_596)
%onnx::Conv_608 = Identity(%onnx::Conv_596)
%onnx::Conv_605 = Identity(%onnx::Conv_596)
%onnx::Conv_602 = Identity(%onnx::Conv_596)
%onnx::Conv_599 = Identity(%onnx::Conv_596)
%onnx::Conv_593 = Identity(%onnx::Conv_578)
%onnx::Conv_590 = Identity(%onnx::Conv_578)
%onnx::Conv_587 = Identity(%onnx::Conv_578)
%onnx::Conv_584 = Identity(%onnx::Conv_578)
%onnx::Conv_581 = Identity(%onnx::Conv_578)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_577, %onnx::Conv_578)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_580, %onnx::Conv_581)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_583, %onnx::Conv_584)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_586, %onnx::Conv_587)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_589, %onnx::Conv_590)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_592, %onnx::Conv_593)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_595, %onnx::Conv_596)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_598, %onnx::Conv_599)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_601, %onnx::Conv_602)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_604, %onnx::Conv_605)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_607, %onnx::Conv_608)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_610, %onnx::Conv_611)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_613, %onnx::Conv_614)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%575 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %575
}
|
val_accuracy
| 0 | 57,159,040 | 1,640,500 |
{'zcp_synflow': 73.61865372470578, 'zcp_zen': 61.88322067260742, 'zcp_epe_nas': 7.279341107023081, 'zcp_fisher': 0.07520660012960434, 'zcp_flops': 57159040.0, 'zcp_grad_norm': 17.59659194946289, 'zcp_grasp': -0.02329540252685547, 'zcp_jacov': -16.07320334284376, 'zcp_l2_norm': 557.1242065429688, 'zcp_nwot': 209.4835220317298, 'zcp_params': 1640500.0, 'zcp_plain': -0.00023361660714726895, 'zcp_snip': 30.071922302246094, 'lat_1080ti_1': 0.3882127156978118, 'lat_1080ti_32': 0.2507007956089424, 'lat_1080ti_64': 0.22581159938760867, 'lat_2080ti_1': 0.40772039170114505, 'lat_2080ti_32': 0.29589458745980834, 'lat_2080ti_64': 0.23180083535261897, 'lat_essential_ph_1': 0.33962264150943394, 'lat_eyeriss': 0.315781270056663, 'lat_fpga': 0.26285635124036444, 'lat_gold_6226': 0.2748603076724312, 'lat_gold_6240': 0.3922686085689483, 'lat_pixel2': 0.21739130434782608, 'lat_pixel3': 0.3135462255391097, 'lat_raspi4': 0.32687192847216967, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.15748031496062992, 'lat_silver_4114': 0.45263824864779256, 'lat_silver_4210r': 0.42368216498137523, 'lat_titan_rtx_1': 0.3877671387880719, 'lat_titan_rtx_32': 0.29426538670183433, 'lat_titan_rtx_64': 0.23613237515482285, 'lat_titanx_1': 0.2083932719734901, 'lat_titanx_32': 0.24441143200834292, 'lat_titanx_64': 0.23517229240245932, 'lat_titanxp_1': 0.3662938489153254, 'lat_titanxp_32': 0.26376127079613504, 'lat_titanxp_64': 0.23474775877536252}
| |
FBNet_2968
|
FBNet
|
2968
|
2968
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_568[FLOAT, 16x3x3x3]
%onnx::Conv_569[FLOAT, 16]
%onnx::Conv_571[FLOAT, 48x16x1x1]
%onnx::Conv_572[FLOAT, 48]
%onnx::Conv_574[FLOAT, 48x1x3x3]
%onnx::Conv_577[FLOAT, 24x48x1x1]
%onnx::Conv_578[FLOAT, 24]
%onnx::Conv_580[FLOAT, 144x24x1x1]
%onnx::Conv_581[FLOAT, 144]
%onnx::Conv_583[FLOAT, 144x1x3x3]
%onnx::Conv_586[FLOAT, 24x144x1x1]
%onnx::Conv_589[FLOAT, 72x24x1x1]
%onnx::Conv_590[FLOAT, 72]
%onnx::Conv_592[FLOAT, 72x1x3x3]
%onnx::Conv_595[FLOAT, 24x72x1x1]
%onnx::Conv_598[FLOAT, 24x24x1x1]
%onnx::Conv_601[FLOAT, 24x1x5x5]
%onnx::Conv_604[FLOAT, 24x24x1x1]
%onnx::Conv_607[FLOAT, 72x24x1x1]
%onnx::Conv_610[FLOAT, 72x1x5x5]
%onnx::Conv_613[FLOAT, 32x72x1x1]
%onnx::Conv_614[FLOAT, 32]
%onnx::Conv_616[FLOAT, 96x32x1x1]
%onnx::Conv_617[FLOAT, 96]
%onnx::Conv_619[FLOAT, 96x1x3x3]
%onnx::Conv_622[FLOAT, 32x96x1x1]
%onnx::Conv_625[FLOAT, 96x32x1x1]
%onnx::Conv_628[FLOAT, 96x1x5x5]
%onnx::Conv_631[FLOAT, 64x96x1x1]
%onnx::Conv_632[FLOAT, 64]
%onnx::Conv_634[FLOAT, 384x64x1x1]
%onnx::Conv_635[FLOAT, 384]
%onnx::Conv_637[FLOAT, 384x1x5x5]
%onnx::Conv_640[FLOAT, 64x384x1x1]
%onnx::Conv_643[FLOAT, 64x32x1x1]
%onnx::Conv_646[FLOAT, 64x1x3x3]
%onnx::Conv_649[FLOAT, 64x32x1x1]
%onnx::Conv_652[FLOAT, 384x64x1x1]
%onnx::Conv_655[FLOAT, 384x1x5x5]
%onnx::Conv_658[FLOAT, 64x384x1x1]
%onnx::Conv_661[FLOAT, 64x64x1x1]
%onnx::Conv_664[FLOAT, 64x1x3x3]
%onnx::Conv_667[FLOAT, 112x64x1x1]
%onnx::Conv_668[FLOAT, 112]
%onnx::Conv_670[FLOAT, 112x112x1x1]
%onnx::Conv_673[FLOAT, 112x1x3x3]
%onnx::Conv_676[FLOAT, 112x112x1x1]
%onnx::Conv_679[FLOAT, 672x112x1x1]
%onnx::Conv_680[FLOAT, 672]
%onnx::Conv_682[FLOAT, 672x1x5x5]
%onnx::Conv_685[FLOAT, 112x672x1x1]
%onnx::Conv_688[FLOAT, 112x56x1x1]
%onnx::Conv_691[FLOAT, 112x1x3x3]
%onnx::Conv_694[FLOAT, 112x56x1x1]
%onnx::Conv_697[FLOAT, 672x112x1x1]
%onnx::Conv_700[FLOAT, 672x1x5x5]
%onnx::Conv_703[FLOAT, 184x672x1x1]
%onnx::Conv_704[FLOAT, 184]
%onnx::Conv_706[FLOAT, 184x92x1x1]
%onnx::Conv_709[FLOAT, 184x1x5x5]
%onnx::Conv_712[FLOAT, 184x92x1x1]
%onnx::Conv_715[FLOAT, 1104x184x1x1]
%onnx::Conv_716[FLOAT, 1104]
%onnx::Conv_718[FLOAT, 1104x1x5x5]
%onnx::Conv_721[FLOAT, 184x1104x1x1]
%onnx::Conv_724[FLOAT, 1104x184x1x1]
%onnx::Conv_727[FLOAT, 1104x1x5x5]
%onnx::Conv_730[FLOAT, 352x1104x1x1]
%onnx::Conv_731[FLOAT, 352]
%onnx::Conv_733[FLOAT, 1504x352x1x1]
%onnx::Conv_734[FLOAT, 1504]
) {
%onnx::Conv_728 = Identity(%onnx::Conv_716)
%onnx::Conv_725 = Identity(%onnx::Conv_716)
%onnx::Conv_722 = Identity(%onnx::Conv_704)
%onnx::Conv_719 = Identity(%onnx::Conv_716)
%onnx::Conv_713 = Identity(%onnx::Conv_704)
%onnx::Conv_710 = Identity(%onnx::Conv_704)
%onnx::Conv_707 = Identity(%onnx::Conv_704)
%onnx::Conv_701 = Identity(%onnx::Conv_680)
%onnx::Conv_698 = Identity(%onnx::Conv_680)
%onnx::Conv_695 = Identity(%onnx::Conv_668)
%onnx::Conv_692 = Identity(%onnx::Conv_668)
%onnx::Conv_689 = Identity(%onnx::Conv_668)
%onnx::Conv_686 = Identity(%onnx::Conv_668)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_677 = Identity(%onnx::Conv_668)
%onnx::Conv_674 = Identity(%onnx::Conv_668)
%onnx::Conv_671 = Identity(%onnx::Conv_668)
%onnx::Conv_665 = Identity(%onnx::Conv_632)
%onnx::Conv_662 = Identity(%onnx::Conv_632)
%onnx::Conv_659 = Identity(%onnx::Conv_632)
%onnx::Conv_656 = Identity(%onnx::Conv_635)
%onnx::Conv_653 = Identity(%onnx::Conv_635)
%onnx::Conv_650 = Identity(%onnx::Conv_632)
%onnx::Conv_647 = Identity(%onnx::Conv_632)
%onnx::Conv_644 = Identity(%onnx::Conv_632)
%onnx::Conv_641 = Identity(%onnx::Conv_632)
%onnx::Conv_638 = Identity(%onnx::Conv_635)
%onnx::Conv_629 = Identity(%onnx::Conv_617)
%onnx::Conv_626 = Identity(%onnx::Conv_617)
%onnx::Conv_623 = Identity(%onnx::Conv_614)
%onnx::Conv_620 = Identity(%onnx::Conv_617)
%onnx::Conv_611 = Identity(%onnx::Conv_590)
%onnx::Conv_608 = Identity(%onnx::Conv_590)
%onnx::Conv_605 = Identity(%onnx::Conv_578)
%onnx::Conv_602 = Identity(%onnx::Conv_578)
%onnx::Conv_599 = Identity(%onnx::Conv_578)
%onnx::Conv_596 = Identity(%onnx::Conv_578)
%onnx::Conv_593 = Identity(%onnx::Conv_590)
%onnx::Conv_587 = Identity(%onnx::Conv_578)
%onnx::Conv_584 = Identity(%onnx::Conv_581)
%onnx::Conv_575 = Identity(%onnx::Conv_572)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_568, %onnx::Conv_569)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_571, %onnx::Conv_572)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_574, %onnx::Conv_575)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_577, %onnx::Conv_578)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_580, %onnx::Conv_581)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_583, %onnx::Conv_584)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_586, %onnx::Conv_587)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_589, %onnx::Conv_590)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_592, %onnx::Conv_593)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_595, %onnx::Conv_596)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_598, %onnx::Conv_599)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_601, %onnx::Conv_602)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_604, %onnx::Conv_605)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_607, %onnx::Conv_608)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_610, %onnx::Conv_611)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_613, %onnx::Conv_614)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%566 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %566
}
|
val_accuracy
| 0 | 81,419,648 | 2,397,796 |
{'zcp_synflow': 71.56296918282978, 'zcp_zen': 65.08619689941406, 'zcp_epe_nas': 14.519932073924004, 'zcp_fisher': 0.07087687402963638, 'zcp_flops': 81419648.0, 'zcp_grad_norm': 19.059650421142578, 'zcp_grasp': -0.03170299530029297, 'zcp_jacov': -16.068702986051164, 'zcp_l2_norm': 619.3787841796875, 'zcp_nwot': 213.63575954410712, 'zcp_params': 2397796.0, 'zcp_plain': -0.000576610560528934, 'zcp_snip': 33.26402282714844, 'lat_1080ti_1': 0.3539715154413583, 'lat_1080ti_32': 0.3740836153406378, 'lat_1080ti_64': 0.3879033896240058, 'lat_2080ti_1': 0.3712187177070715, 'lat_2080ti_32': 0.3524235839765623, 'lat_2080ti_64': 0.38775673714792697, 'lat_essential_ph_1': 0.39622641509433965, 'lat_eyeriss': 0.557423945134139, 'lat_fpga': 0.6188365212400677, 'lat_gold_6226': 0.5027791546378616, 'lat_gold_6240': 0.5918315022423635, 'lat_pixel2': 0.5652173913043478, 'lat_pixel3': 0.5427227132380912, 'lat_raspi4': 0.5921564145070863, 'lat_samsung_a50': 0.3157894736842105, 'lat_samsung_s7': 0.2125984251968504, 'lat_silver_4114': 0.5490705965572577, 'lat_silver_4210r': 0.5274344425990383, 'lat_titan_rtx_1': 0.35772340139588926, 'lat_titan_rtx_32': 0.32905027181868884, 'lat_titan_rtx_64': 0.34571300589775517, 'lat_titanx_1': 0.19557742421099414, 'lat_titanx_32': 0.32876020003998907, 'lat_titanx_64': 0.40513131851667433, 'lat_titanxp_1': 0.3662996185454729, 'lat_titanxp_32': 0.3273447210133329, 'lat_titanxp_64': 0.3851888353259407}
| |
FBNet_3244
|
FBNet
|
3244
|
3244
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_558[FLOAT, 16x3x3x3]
%onnx::Conv_559[FLOAT, 16]
%onnx::Conv_561[FLOAT, 96x16x1x1]
%onnx::Conv_562[FLOAT, 96]
%onnx::Conv_564[FLOAT, 96x1x5x5]
%onnx::Conv_567[FLOAT, 16x96x1x1]
%onnx::Conv_570[FLOAT, 16x16x1x1]
%onnx::Conv_573[FLOAT, 16x1x3x3]
%onnx::Conv_576[FLOAT, 24x16x1x1]
%onnx::Conv_577[FLOAT, 24]
%onnx::Conv_579[FLOAT, 24x24x1x1]
%onnx::Conv_582[FLOAT, 24x1x5x5]
%onnx::Conv_585[FLOAT, 24x24x1x1]
%onnx::Conv_588[FLOAT, 144x24x1x1]
%onnx::Conv_589[FLOAT, 144]
%onnx::Conv_591[FLOAT, 144x1x3x3]
%onnx::Conv_594[FLOAT, 24x144x1x1]
%onnx::Conv_597[FLOAT, 144x24x1x1]
%onnx::Conv_600[FLOAT, 144x1x5x5]
%onnx::Conv_603[FLOAT, 32x144x1x1]
%onnx::Conv_604[FLOAT, 32]
%onnx::Conv_606[FLOAT, 32x32x1x1]
%onnx::Conv_609[FLOAT, 32x1x5x5]
%onnx::Conv_612[FLOAT, 32x32x1x1]
%onnx::Conv_615[FLOAT, 32x32x1x1]
%onnx::Conv_618[FLOAT, 32x1x5x5]
%onnx::Conv_621[FLOAT, 32x32x1x1]
%onnx::Conv_624[FLOAT, 96x32x1x1]
%onnx::Conv_627[FLOAT, 96x1x3x3]
%onnx::Conv_630[FLOAT, 64x96x1x1]
%onnx::Conv_631[FLOAT, 64]
%onnx::Conv_633[FLOAT, 64x64x1x1]
%onnx::Conv_636[FLOAT, 64x1x3x3]
%onnx::Conv_639[FLOAT, 64x64x1x1]
%onnx::Conv_642[FLOAT, 64x64x1x1]
%onnx::Conv_645[FLOAT, 64x1x5x5]
%onnx::Conv_648[FLOAT, 64x64x1x1]
%onnx::Conv_651[FLOAT, 112x64x1x1]
%onnx::Conv_652[FLOAT, 112]
%onnx::Conv_654[FLOAT, 112x56x1x1]
%onnx::Conv_657[FLOAT, 112x1x3x3]
%onnx::Conv_660[FLOAT, 112x56x1x1]
%onnx::Conv_663[FLOAT, 336x112x1x1]
%onnx::Conv_664[FLOAT, 336]
%onnx::Conv_666[FLOAT, 336x1x3x3]
%onnx::Conv_669[FLOAT, 112x336x1x1]
%onnx::Conv_672[FLOAT, 112x56x1x1]
%onnx::Conv_675[FLOAT, 112x1x3x3]
%onnx::Conv_678[FLOAT, 112x56x1x1]
%onnx::Conv_681[FLOAT, 336x112x1x1]
%onnx::Conv_684[FLOAT, 336x1x5x5]
%onnx::Conv_687[FLOAT, 184x336x1x1]
%onnx::Conv_688[FLOAT, 184]
%onnx::Conv_690[FLOAT, 184x184x1x1]
%onnx::Conv_693[FLOAT, 184x1x3x3]
%onnx::Conv_696[FLOAT, 184x184x1x1]
%onnx::Conv_699[FLOAT, 552x184x1x1]
%onnx::Conv_700[FLOAT, 552]
%onnx::Conv_702[FLOAT, 552x1x3x3]
%onnx::Conv_705[FLOAT, 184x552x1x1]
%onnx::Conv_708[FLOAT, 1104x184x1x1]
%onnx::Conv_709[FLOAT, 1104]
%onnx::Conv_711[FLOAT, 1104x1x3x3]
%onnx::Conv_714[FLOAT, 184x1104x1x1]
%onnx::Conv_717[FLOAT, 1104x184x1x1]
%onnx::Conv_720[FLOAT, 1104x1x3x3]
%onnx::Conv_723[FLOAT, 352x1104x1x1]
%onnx::Conv_724[FLOAT, 352]
%onnx::Conv_726[FLOAT, 1504x352x1x1]
%onnx::Conv_727[FLOAT, 1504]
) {
%onnx::Conv_721 = Identity(%onnx::Conv_709)
%onnx::Conv_718 = Identity(%onnx::Conv_709)
%onnx::Conv_715 = Identity(%onnx::Conv_688)
%onnx::Conv_712 = Identity(%onnx::Conv_709)
%onnx::Conv_706 = Identity(%onnx::Conv_688)
%onnx::Conv_703 = Identity(%onnx::Conv_700)
%onnx::Conv_697 = Identity(%onnx::Conv_688)
%onnx::Conv_694 = Identity(%onnx::Conv_688)
%onnx::Conv_691 = Identity(%onnx::Conv_688)
%onnx::Conv_685 = Identity(%onnx::Conv_664)
%onnx::Conv_682 = Identity(%onnx::Conv_664)
%onnx::Conv_679 = Identity(%onnx::Conv_652)
%onnx::Conv_676 = Identity(%onnx::Conv_652)
%onnx::Conv_673 = Identity(%onnx::Conv_652)
%onnx::Conv_670 = Identity(%onnx::Conv_652)
%onnx::Conv_667 = Identity(%onnx::Conv_664)
%onnx::Conv_661 = Identity(%onnx::Conv_652)
%onnx::Conv_658 = Identity(%onnx::Conv_652)
%onnx::Conv_655 = Identity(%onnx::Conv_652)
%onnx::Conv_649 = Identity(%onnx::Conv_631)
%onnx::Conv_646 = Identity(%onnx::Conv_631)
%onnx::Conv_643 = Identity(%onnx::Conv_631)
%onnx::Conv_640 = Identity(%onnx::Conv_631)
%onnx::Conv_637 = Identity(%onnx::Conv_631)
%onnx::Conv_634 = Identity(%onnx::Conv_631)
%onnx::Conv_628 = Identity(%onnx::Conv_562)
%onnx::Conv_625 = Identity(%onnx::Conv_562)
%onnx::Conv_622 = Identity(%onnx::Conv_604)
%onnx::Conv_619 = Identity(%onnx::Conv_604)
%onnx::Conv_616 = Identity(%onnx::Conv_604)
%onnx::Conv_613 = Identity(%onnx::Conv_604)
%onnx::Conv_610 = Identity(%onnx::Conv_604)
%onnx::Conv_607 = Identity(%onnx::Conv_604)
%onnx::Conv_601 = Identity(%onnx::Conv_589)
%onnx::Conv_598 = Identity(%onnx::Conv_589)
%onnx::Conv_595 = Identity(%onnx::Conv_577)
%onnx::Conv_592 = Identity(%onnx::Conv_589)
%onnx::Conv_586 = Identity(%onnx::Conv_577)
%onnx::Conv_583 = Identity(%onnx::Conv_577)
%onnx::Conv_580 = Identity(%onnx::Conv_577)
%onnx::Conv_574 = Identity(%onnx::Conv_559)
%onnx::Conv_571 = Identity(%onnx::Conv_559)
%onnx::Conv_568 = Identity(%onnx::Conv_559)
%onnx::Conv_565 = Identity(%onnx::Conv_562)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_558, %onnx::Conv_559)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_561, %onnx::Conv_562)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_564, %onnx::Conv_565)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_567, %onnx::Conv_568)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_570, %onnx::Conv_571)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_573, %onnx::Conv_574)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_576, %onnx::Conv_577)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_579, %onnx::Conv_580)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_582, %onnx::Conv_583)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_585, %onnx::Conv_586)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_588, %onnx::Conv_589)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_591, %onnx::Conv_592)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_594, %onnx::Conv_595)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_597, %onnx::Conv_598)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_600, %onnx::Conv_601)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_603, %onnx::Conv_604)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_606, %onnx::Conv_607)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_609, %onnx::Conv_610)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_612, %onnx::Conv_613)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_615, %onnx::Conv_616)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%556 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %556
}
|
val_accuracy
| 0 | 69,653,376 | 2,283,404 |
{'zcp_synflow': 74.8164286550526, 'zcp_zen': 63.879722595214844, 'zcp_epe_nas': 10.769795899000204, 'zcp_fisher': 0.11664873361587524, 'zcp_flops': 69653376.0, 'zcp_grad_norm': 21.67709732055664, 'zcp_grasp': -0.13941574096679688, 'zcp_jacov': -16.05773892536145, 'zcp_l2_norm': 604.85302734375, 'zcp_nwot': 212.6968392136315, 'zcp_params': 2283404.0, 'zcp_plain': 0.0024125277996063232, 'zcp_snip': 40.80100631713867, 'lat_1080ti_1': 0.3736899760607106, 'lat_1080ti_32': 0.4766835753048685, 'lat_1080ti_64': 0.4150414107175138, 'lat_2080ti_1': 0.40092789641335314, 'lat_2080ti_32': 0.4353796464802201, 'lat_2080ti_64': 0.4390987351156328, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.452322446958722, 'lat_fpga': 0.4693693782839726, 'lat_gold_6226': 0.3640755301121627, 'lat_gold_6240': 0.42757083622172065, 'lat_pixel2': 0.45652173913043476, 'lat_pixel3': 0.42831693998985326, 'lat_raspi4': 0.5459621812082646, 'lat_samsung_a50': 0.2, 'lat_samsung_s7': 0.11023622047244094, 'lat_silver_4114': 0.43849529980782675, 'lat_silver_4210r': 0.43321423487360544, 'lat_titan_rtx_1': 0.37276415810432195, 'lat_titan_rtx_32': 0.40360692848854984, 'lat_titan_rtx_64': 0.44155399191575123, 'lat_titanx_1': 0.20047626471448224, 'lat_titanx_32': 0.416929078613807, 'lat_titanx_64': 0.4523027580827765, 'lat_titanxp_1': 0.386315299285969, 'lat_titanxp_32': 0.4275344486936511, 'lat_titanxp_64': 0.43923198642937444}
| |
FBNet_1609
|
FBNet
|
1609
|
1609
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_642[FLOAT, 16x3x3x3]
%onnx::Conv_643[FLOAT, 16]
%onnx::Conv_645[FLOAT, 16x16x1x1]
%onnx::Conv_648[FLOAT, 16x1x5x5]
%onnx::Conv_651[FLOAT, 16x16x1x1]
%onnx::Conv_654[FLOAT, 96x16x1x1]
%onnx::Conv_655[FLOAT, 96]
%onnx::Conv_657[FLOAT, 96x1x3x3]
%onnx::Conv_660[FLOAT, 24x96x1x1]
%onnx::Conv_661[FLOAT, 24]
%onnx::Conv_663[FLOAT, 24x12x1x1]
%onnx::Conv_666[FLOAT, 24x1x5x5]
%onnx::Conv_669[FLOAT, 24x12x1x1]
%onnx::Conv_672[FLOAT, 24x12x1x1]
%onnx::Conv_675[FLOAT, 24x1x3x3]
%onnx::Conv_678[FLOAT, 24x12x1x1]
%onnx::Conv_681[FLOAT, 32x24x1x1]
%onnx::Conv_682[FLOAT, 32]
%onnx::Conv_684[FLOAT, 192x32x1x1]
%onnx::Conv_685[FLOAT, 192]
%onnx::Conv_687[FLOAT, 192x1x5x5]
%onnx::Conv_690[FLOAT, 32x192x1x1]
%onnx::Conv_693[FLOAT, 192x32x1x1]
%onnx::Conv_696[FLOAT, 192x1x3x3]
%onnx::Conv_699[FLOAT, 32x192x1x1]
%onnx::Conv_702[FLOAT, 32x16x1x1]
%onnx::Conv_705[FLOAT, 32x1x5x5]
%onnx::Conv_708[FLOAT, 64x16x1x1]
%onnx::Conv_709[FLOAT, 64]
%onnx::Conv_711[FLOAT, 384x64x1x1]
%onnx::Conv_712[FLOAT, 384]
%onnx::Conv_714[FLOAT, 384x1x3x3]
%onnx::Conv_717[FLOAT, 64x384x1x1]
%onnx::Conv_720[FLOAT, 64x64x1x1]
%onnx::Conv_723[FLOAT, 64x1x5x5]
%onnx::Conv_726[FLOAT, 64x64x1x1]
%onnx::Conv_729[FLOAT, 192x64x1x1]
%onnx::Conv_732[FLOAT, 192x1x5x5]
%onnx::Conv_735[FLOAT, 64x192x1x1]
%onnx::Conv_738[FLOAT, 64x32x1x1]
%onnx::Conv_741[FLOAT, 64x1x3x3]
%onnx::Conv_744[FLOAT, 112x32x1x1]
%onnx::Conv_745[FLOAT, 112]
%onnx::Conv_747[FLOAT, 672x112x1x1]
%onnx::Conv_748[FLOAT, 672]
%onnx::Conv_750[FLOAT, 672x1x3x3]
%onnx::Conv_753[FLOAT, 112x672x1x1]
%onnx::Conv_756[FLOAT, 672x112x1x1]
%onnx::Conv_759[FLOAT, 672x1x3x3]
%onnx::Conv_762[FLOAT, 112x672x1x1]
%onnx::Conv_765[FLOAT, 336x112x1x1]
%onnx::Conv_766[FLOAT, 336]
%onnx::Conv_768[FLOAT, 336x1x3x3]
%onnx::Conv_771[FLOAT, 112x336x1x1]
%onnx::Conv_774[FLOAT, 112x56x1x1]
%onnx::Conv_777[FLOAT, 112x1x3x3]
%onnx::Conv_780[FLOAT, 184x56x1x1]
%onnx::Conv_781[FLOAT, 184]
%onnx::Conv_783[FLOAT, 1104x184x1x1]
%onnx::Conv_784[FLOAT, 1104]
%onnx::Conv_786[FLOAT, 1104x1x3x3]
%onnx::Conv_789[FLOAT, 184x1104x1x1]
%onnx::Conv_792[FLOAT, 552x184x1x1]
%onnx::Conv_793[FLOAT, 552]
%onnx::Conv_795[FLOAT, 552x1x3x3]
%onnx::Conv_798[FLOAT, 184x552x1x1]
%onnx::Conv_801[FLOAT, 1104x184x1x1]
%onnx::Conv_804[FLOAT, 1104x1x3x3]
%onnx::Conv_807[FLOAT, 184x1104x1x1]
%onnx::Conv_810[FLOAT, 184x184x1x1]
%onnx::Conv_813[FLOAT, 184x1x5x5]
%onnx::Conv_816[FLOAT, 352x184x1x1]
%onnx::Conv_817[FLOAT, 352]
%onnx::Conv_819[FLOAT, 1504x352x1x1]
%onnx::Conv_820[FLOAT, 1504]
) {
%onnx::Conv_814 = Identity(%onnx::Conv_781)
%onnx::Conv_811 = Identity(%onnx::Conv_781)
%onnx::Conv_808 = Identity(%onnx::Conv_781)
%onnx::Conv_805 = Identity(%onnx::Conv_784)
%onnx::Conv_802 = Identity(%onnx::Conv_784)
%onnx::Conv_799 = Identity(%onnx::Conv_781)
%onnx::Conv_796 = Identity(%onnx::Conv_793)
%onnx::Conv_790 = Identity(%onnx::Conv_781)
%onnx::Conv_787 = Identity(%onnx::Conv_784)
%onnx::Conv_778 = Identity(%onnx::Conv_745)
%onnx::Conv_775 = Identity(%onnx::Conv_745)
%onnx::Conv_772 = Identity(%onnx::Conv_745)
%onnx::Conv_769 = Identity(%onnx::Conv_766)
%onnx::Conv_763 = Identity(%onnx::Conv_745)
%onnx::Conv_760 = Identity(%onnx::Conv_748)
%onnx::Conv_757 = Identity(%onnx::Conv_748)
%onnx::Conv_754 = Identity(%onnx::Conv_745)
%onnx::Conv_751 = Identity(%onnx::Conv_748)
%onnx::Conv_742 = Identity(%onnx::Conv_709)
%onnx::Conv_739 = Identity(%onnx::Conv_709)
%onnx::Conv_736 = Identity(%onnx::Conv_709)
%onnx::Conv_733 = Identity(%onnx::Conv_685)
%onnx::Conv_730 = Identity(%onnx::Conv_685)
%onnx::Conv_727 = Identity(%onnx::Conv_709)
%onnx::Conv_724 = Identity(%onnx::Conv_709)
%onnx::Conv_721 = Identity(%onnx::Conv_709)
%onnx::Conv_718 = Identity(%onnx::Conv_709)
%onnx::Conv_715 = Identity(%onnx::Conv_712)
%onnx::Conv_706 = Identity(%onnx::Conv_682)
%onnx::Conv_703 = Identity(%onnx::Conv_682)
%onnx::Conv_700 = Identity(%onnx::Conv_682)
%onnx::Conv_697 = Identity(%onnx::Conv_685)
%onnx::Conv_694 = Identity(%onnx::Conv_685)
%onnx::Conv_691 = Identity(%onnx::Conv_682)
%onnx::Conv_688 = Identity(%onnx::Conv_685)
%onnx::Conv_679 = Identity(%onnx::Conv_661)
%onnx::Conv_676 = Identity(%onnx::Conv_661)
%onnx::Conv_673 = Identity(%onnx::Conv_661)
%onnx::Conv_670 = Identity(%onnx::Conv_661)
%onnx::Conv_667 = Identity(%onnx::Conv_661)
%onnx::Conv_664 = Identity(%onnx::Conv_661)
%onnx::Conv_658 = Identity(%onnx::Conv_655)
%onnx::Conv_652 = Identity(%onnx::Conv_643)
%onnx::Conv_649 = Identity(%onnx::Conv_643)
%onnx::Conv_646 = Identity(%onnx::Conv_643)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_642, %onnx::Conv_643)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.3/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.5/relu/Relu_output_0 = Relu(%/cells.5/conv/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/relu/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/relu/Relu_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%640 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %640
}
|
val_accuracy
| 0 | 78,301,056 | 2,403,940 |
{'zcp_synflow': 73.78564601694413, 'zcp_zen': 67.82715606689453, 'zcp_epe_nas': 8.546049997607733, 'zcp_fisher': 0.14203907549381256, 'zcp_flops': 78301056.0, 'zcp_grad_norm': 23.581666946411133, 'zcp_grasp': -0.1331920623779297, 'zcp_jacov': -16.055871154781933, 'zcp_l2_norm': 674.8174438476562, 'zcp_nwot': 211.11332257707716, 'zcp_params': 2403940.0, 'zcp_plain': -0.002069745445623994, 'zcp_snip': 43.067901611328125, 'lat_1080ti_1': 0.5494041607238304, 'lat_1080ti_32': 0.4529847045094547, 'lat_1080ti_64': 0.3390389377330411, 'lat_2080ti_1': 0.5651226316369473, 'lat_2080ti_32': 0.43420969966046746, 'lat_2080ti_64': 0.34904491835551066, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.5257275411218894, 'lat_fpga': 0.6768818215102067, 'lat_gold_6226': 0.5336762294770726, 'lat_gold_6240': 0.6943238219250681, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.4765209711467244, 'lat_raspi4': 0.5041977608005933, 'lat_samsung_a50': 0.24210526315789474, 'lat_samsung_s7': 0.2283464566929134, 'lat_silver_4114': 0.7091817671023217, 'lat_silver_4210r': 0.69701747779472, 'lat_titan_rtx_1': 0.5279731135734608, 'lat_titan_rtx_32': 0.45221471188414686, 'lat_titan_rtx_64': 0.36423688032949786, 'lat_titanx_1': 0.2879504521492509, 'lat_titanx_32': 0.37579069323697845, 'lat_titanx_64': 0.34546650831529213, 'lat_titanxp_1': 0.5071090646070641, 'lat_titanxp_32': 0.41521152896427077, 'lat_titanxp_64': 0.33160959419485236}
| |
FBNet_2927
|
FBNet
|
2927
|
2927
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_632[FLOAT, 16x3x3x3]
%onnx::Conv_633[FLOAT, 16]
%onnx::Conv_635[FLOAT, 16x8x1x1]
%onnx::Conv_638[FLOAT, 16x1x5x5]
%onnx::Conv_641[FLOAT, 16x8x1x1]
%onnx::Conv_644[FLOAT, 96x16x1x1]
%onnx::Conv_645[FLOAT, 96]
%onnx::Conv_647[FLOAT, 96x1x5x5]
%onnx::Conv_650[FLOAT, 24x96x1x1]
%onnx::Conv_651[FLOAT, 24]
%onnx::Conv_653[FLOAT, 24x12x1x1]
%onnx::Conv_656[FLOAT, 24x1x5x5]
%onnx::Conv_659[FLOAT, 24x12x1x1]
%onnx::Conv_662[FLOAT, 24x24x1x1]
%onnx::Conv_665[FLOAT, 24x1x3x3]
%onnx::Conv_668[FLOAT, 24x24x1x1]
%onnx::Conv_671[FLOAT, 144x24x1x1]
%onnx::Conv_672[FLOAT, 144]
%onnx::Conv_674[FLOAT, 144x1x5x5]
%onnx::Conv_677[FLOAT, 32x144x1x1]
%onnx::Conv_678[FLOAT, 32]
%onnx::Conv_680[FLOAT, 32x16x1x1]
%onnx::Conv_683[FLOAT, 32x1x5x5]
%onnx::Conv_686[FLOAT, 32x16x1x1]
%onnx::Conv_689[FLOAT, 32x16x1x1]
%onnx::Conv_692[FLOAT, 32x1x3x3]
%onnx::Conv_695[FLOAT, 64x16x1x1]
%onnx::Conv_696[FLOAT, 64]
%onnx::Conv_698[FLOAT, 64x64x1x1]
%onnx::Conv_701[FLOAT, 64x1x3x3]
%onnx::Conv_704[FLOAT, 64x64x1x1]
%onnx::Conv_707[FLOAT, 192x64x1x1]
%onnx::Conv_708[FLOAT, 192]
%onnx::Conv_710[FLOAT, 192x1x5x5]
%onnx::Conv_713[FLOAT, 64x192x1x1]
%onnx::Conv_716[FLOAT, 64x32x1x1]
%onnx::Conv_719[FLOAT, 64x1x3x3]
%onnx::Conv_722[FLOAT, 64x32x1x1]
%onnx::Conv_725[FLOAT, 64x64x1x1]
%onnx::Conv_728[FLOAT, 64x1x5x5]
%onnx::Conv_731[FLOAT, 112x64x1x1]
%onnx::Conv_732[FLOAT, 112]
%onnx::Conv_734[FLOAT, 672x112x1x1]
%onnx::Conv_735[FLOAT, 672]
%onnx::Conv_737[FLOAT, 672x1x5x5]
%onnx::Conv_740[FLOAT, 112x672x1x1]
%onnx::Conv_743[FLOAT, 336x112x1x1]
%onnx::Conv_744[FLOAT, 336]
%onnx::Conv_746[FLOAT, 336x1x3x3]
%onnx::Conv_749[FLOAT, 112x336x1x1]
%onnx::Conv_752[FLOAT, 336x112x1x1]
%onnx::Conv_755[FLOAT, 336x1x3x3]
%onnx::Conv_758[FLOAT, 112x336x1x1]
%onnx::Conv_761[FLOAT, 336x112x1x1]
%onnx::Conv_764[FLOAT, 336x1x5x5]
%onnx::Conv_767[FLOAT, 184x336x1x1]
%onnx::Conv_768[FLOAT, 184]
%onnx::Conv_770[FLOAT, 184x184x1x1]
%onnx::Conv_773[FLOAT, 184x1x3x3]
%onnx::Conv_776[FLOAT, 184x184x1x1]
%onnx::Conv_779[FLOAT, 184x184x1x1]
%onnx::Conv_782[FLOAT, 184x1x3x3]
%onnx::Conv_785[FLOAT, 184x184x1x1]
%onnx::Conv_788[FLOAT, 1104x184x1x1]
%onnx::Conv_789[FLOAT, 1104]
%onnx::Conv_791[FLOAT, 1104x1x3x3]
%onnx::Conv_794[FLOAT, 184x1104x1x1]
%onnx::Conv_797[FLOAT, 552x184x1x1]
%onnx::Conv_798[FLOAT, 552]
%onnx::Conv_800[FLOAT, 552x1x5x5]
%onnx::Conv_803[FLOAT, 352x552x1x1]
%onnx::Conv_804[FLOAT, 352]
%onnx::Conv_806[FLOAT, 1504x352x1x1]
%onnx::Conv_807[FLOAT, 1504]
) {
%onnx::Conv_801 = Identity(%onnx::Conv_798)
%onnx::Conv_795 = Identity(%onnx::Conv_768)
%onnx::Conv_792 = Identity(%onnx::Conv_789)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_768)
%onnx::Conv_780 = Identity(%onnx::Conv_768)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_768)
%onnx::Conv_771 = Identity(%onnx::Conv_768)
%onnx::Conv_765 = Identity(%onnx::Conv_744)
%onnx::Conv_762 = Identity(%onnx::Conv_744)
%onnx::Conv_759 = Identity(%onnx::Conv_732)
%onnx::Conv_756 = Identity(%onnx::Conv_744)
%onnx::Conv_753 = Identity(%onnx::Conv_744)
%onnx::Conv_750 = Identity(%onnx::Conv_732)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_729 = Identity(%onnx::Conv_696)
%onnx::Conv_726 = Identity(%onnx::Conv_696)
%onnx::Conv_723 = Identity(%onnx::Conv_696)
%onnx::Conv_720 = Identity(%onnx::Conv_696)
%onnx::Conv_717 = Identity(%onnx::Conv_696)
%onnx::Conv_714 = Identity(%onnx::Conv_696)
%onnx::Conv_711 = Identity(%onnx::Conv_708)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_696)
%onnx::Conv_699 = Identity(%onnx::Conv_696)
%onnx::Conv_693 = Identity(%onnx::Conv_678)
%onnx::Conv_690 = Identity(%onnx::Conv_678)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_678)
%onnx::Conv_681 = Identity(%onnx::Conv_678)
%onnx::Conv_675 = Identity(%onnx::Conv_672)
%onnx::Conv_669 = Identity(%onnx::Conv_651)
%onnx::Conv_666 = Identity(%onnx::Conv_651)
%onnx::Conv_663 = Identity(%onnx::Conv_651)
%onnx::Conv_660 = Identity(%onnx::Conv_651)
%onnx::Conv_657 = Identity(%onnx::Conv_651)
%onnx::Conv_654 = Identity(%onnx::Conv_651)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%onnx::Conv_642 = Identity(%onnx::Conv_633)
%onnx::Conv_639 = Identity(%onnx::Conv_633)
%onnx::Conv_636 = Identity(%onnx::Conv_633)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_632, %onnx::Conv_633)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%630 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %630
}
|
val_accuracy
| 0 | 69,256,832 | 2,081,900 |
{'zcp_synflow': 73.44444842995286, 'zcp_zen': 65.56463623046875, 'zcp_epe_nas': 24.30957718378181, 'zcp_fisher': 0.08334095031023026, 'zcp_flops': 69256832.0, 'zcp_grad_norm': 22.70789337158203, 'zcp_grasp': -0.04132843017578125, 'zcp_jacov': -16.055147676994032, 'zcp_l2_norm': 615.8256225585938, 'zcp_nwot': 209.69910879493818, 'zcp_params': 2081900.0, 'zcp_plain': -0.001773883355781436, 'zcp_snip': 36.7384147644043, 'lat_1080ti_1': 0.5109770521326301, 'lat_1080ti_32': 0.42770069721644116, 'lat_1080ti_64': 0.35767871633732434, 'lat_2080ti_1': 0.5206792998000972, 'lat_2080ti_32': 0.4838544277436129, 'lat_2080ti_64': 0.3800390461049232, 'lat_essential_ph_1': 0.3018867924528302, 'lat_eyeriss': 0.3955861588396017, 'lat_fpga': 0.45785135415945133, 'lat_gold_6226': 0.3857667965197784, 'lat_gold_6240': 0.5017071834403319, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.4373281881548904, 'lat_raspi4': 0.5147539733428229, 'lat_samsung_a50': 0.18947368421052632, 'lat_samsung_s7': 0.16535433070866143, 'lat_silver_4114': 0.5172797925242955, 'lat_silver_4210r': 0.49050987208078006, 'lat_titan_rtx_1': 0.5066227648181905, 'lat_titan_rtx_32': 0.46901419004459816, 'lat_titan_rtx_64': 0.3983048836037859, 'lat_titanx_1': 0.26963643723046415, 'lat_titanx_32': 0.41154623973604415, 'lat_titanx_64': 0.372029066420013, 'lat_titanxp_1': 0.4692089891604591, 'lat_titanxp_32': 0.4338756055904485, 'lat_titanxp_64': 0.39807743022307984}
| |
FBNet_1791
|
FBNet
|
1791
|
1791
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_623[FLOAT, 16x3x3x3]
%onnx::Conv_624[FLOAT, 16]
%onnx::Conv_626[FLOAT, 16x8x1x1]
%onnx::Conv_629[FLOAT, 16x1x3x3]
%onnx::Conv_632[FLOAT, 16x8x1x1]
%onnx::Conv_635[FLOAT, 48x16x1x1]
%onnx::Conv_636[FLOAT, 48]
%onnx::Conv_638[FLOAT, 48x1x3x3]
%onnx::Conv_641[FLOAT, 24x48x1x1]
%onnx::Conv_642[FLOAT, 24]
%onnx::Conv_644[FLOAT, 24x12x1x1]
%onnx::Conv_647[FLOAT, 24x1x5x5]
%onnx::Conv_650[FLOAT, 24x12x1x1]
%onnx::Conv_653[FLOAT, 24x12x1x1]
%onnx::Conv_656[FLOAT, 24x1x3x3]
%onnx::Conv_659[FLOAT, 24x12x1x1]
%onnx::Conv_662[FLOAT, 144x24x1x1]
%onnx::Conv_663[FLOAT, 144]
%onnx::Conv_665[FLOAT, 144x1x5x5]
%onnx::Conv_668[FLOAT, 32x144x1x1]
%onnx::Conv_669[FLOAT, 32]
%onnx::Conv_671[FLOAT, 32x16x1x1]
%onnx::Conv_674[FLOAT, 32x1x3x3]
%onnx::Conv_677[FLOAT, 32x16x1x1]
%onnx::Conv_680[FLOAT, 32x16x1x1]
%onnx::Conv_683[FLOAT, 32x1x5x5]
%onnx::Conv_686[FLOAT, 32x16x1x1]
%onnx::Conv_689[FLOAT, 32x32x1x1]
%onnx::Conv_692[FLOAT, 32x1x3x3]
%onnx::Conv_695[FLOAT, 64x32x1x1]
%onnx::Conv_696[FLOAT, 64]
%onnx::Conv_698[FLOAT, 384x64x1x1]
%onnx::Conv_699[FLOAT, 384]
%onnx::Conv_701[FLOAT, 384x1x5x5]
%onnx::Conv_704[FLOAT, 64x384x1x1]
%onnx::Conv_707[FLOAT, 192x64x1x1]
%onnx::Conv_708[FLOAT, 192]
%onnx::Conv_710[FLOAT, 192x1x3x3]
%onnx::Conv_713[FLOAT, 64x192x1x1]
%onnx::Conv_716[FLOAT, 192x64x1x1]
%onnx::Conv_719[FLOAT, 192x1x5x5]
%onnx::Conv_722[FLOAT, 64x192x1x1]
%onnx::Conv_725[FLOAT, 192x64x1x1]
%onnx::Conv_728[FLOAT, 192x1x3x3]
%onnx::Conv_731[FLOAT, 112x192x1x1]
%onnx::Conv_732[FLOAT, 112]
%onnx::Conv_734[FLOAT, 112x56x1x1]
%onnx::Conv_737[FLOAT, 112x1x5x5]
%onnx::Conv_740[FLOAT, 112x56x1x1]
%onnx::Conv_743[FLOAT, 672x112x1x1]
%onnx::Conv_744[FLOAT, 672]
%onnx::Conv_746[FLOAT, 672x1x3x3]
%onnx::Conv_749[FLOAT, 184x672x1x1]
%onnx::Conv_750[FLOAT, 184]
%onnx::Conv_752[FLOAT, 552x184x1x1]
%onnx::Conv_753[FLOAT, 552]
%onnx::Conv_755[FLOAT, 552x1x5x5]
%onnx::Conv_758[FLOAT, 184x552x1x1]
%onnx::Conv_761[FLOAT, 552x184x1x1]
%onnx::Conv_764[FLOAT, 552x1x3x3]
%onnx::Conv_767[FLOAT, 184x552x1x1]
%onnx::Conv_770[FLOAT, 184x184x1x1]
%onnx::Conv_773[FLOAT, 184x1x3x3]
%onnx::Conv_776[FLOAT, 184x184x1x1]
%onnx::Conv_779[FLOAT, 1104x184x1x1]
%onnx::Conv_780[FLOAT, 1104]
%onnx::Conv_782[FLOAT, 1104x1x3x3]
%onnx::Conv_785[FLOAT, 352x1104x1x1]
%onnx::Conv_786[FLOAT, 352]
%onnx::Conv_788[FLOAT, 1504x352x1x1]
%onnx::Conv_789[FLOAT, 1504]
) {
%onnx::Conv_783 = Identity(%onnx::Conv_780)
%onnx::Conv_777 = Identity(%onnx::Conv_750)
%onnx::Conv_774 = Identity(%onnx::Conv_750)
%onnx::Conv_771 = Identity(%onnx::Conv_750)
%onnx::Conv_768 = Identity(%onnx::Conv_750)
%onnx::Conv_765 = Identity(%onnx::Conv_753)
%onnx::Conv_762 = Identity(%onnx::Conv_753)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_732)
%onnx::Conv_735 = Identity(%onnx::Conv_732)
%onnx::Conv_729 = Identity(%onnx::Conv_708)
%onnx::Conv_726 = Identity(%onnx::Conv_708)
%onnx::Conv_723 = Identity(%onnx::Conv_696)
%onnx::Conv_720 = Identity(%onnx::Conv_708)
%onnx::Conv_717 = Identity(%onnx::Conv_708)
%onnx::Conv_714 = Identity(%onnx::Conv_696)
%onnx::Conv_711 = Identity(%onnx::Conv_708)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%onnx::Conv_693 = Identity(%onnx::Conv_669)
%onnx::Conv_690 = Identity(%onnx::Conv_669)
%onnx::Conv_687 = Identity(%onnx::Conv_669)
%onnx::Conv_684 = Identity(%onnx::Conv_669)
%onnx::Conv_681 = Identity(%onnx::Conv_669)
%onnx::Conv_678 = Identity(%onnx::Conv_669)
%onnx::Conv_675 = Identity(%onnx::Conv_669)
%onnx::Conv_672 = Identity(%onnx::Conv_669)
%onnx::Conv_666 = Identity(%onnx::Conv_663)
%onnx::Conv_660 = Identity(%onnx::Conv_642)
%onnx::Conv_657 = Identity(%onnx::Conv_642)
%onnx::Conv_654 = Identity(%onnx::Conv_642)
%onnx::Conv_651 = Identity(%onnx::Conv_642)
%onnx::Conv_648 = Identity(%onnx::Conv_642)
%onnx::Conv_645 = Identity(%onnx::Conv_642)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%onnx::Conv_633 = Identity(%onnx::Conv_624)
%onnx::Conv_630 = Identity(%onnx::Conv_624)
%onnx::Conv_627 = Identity(%onnx::Conv_624)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_623, %onnx::Conv_624)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%621 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %621
}
|
val_accuracy
| 0 | 57,966,720 | 2,194,060 |
{'zcp_synflow': 66.1108100491401, 'zcp_zen': 60.25164031982422, 'zcp_epe_nas': 12.608884823075153, 'zcp_fisher': 0.053423333913087845, 'zcp_flops': 57966720.0, 'zcp_grad_norm': 18.496891021728516, 'zcp_grasp': -0.025694847106933594, 'zcp_jacov': -16.071882888993855, 'zcp_l2_norm': 574.891845703125, 'zcp_nwot': 207.35306895479513, 'zcp_params': 2194060.0, 'zcp_plain': 0.004142379388213158, 'zcp_snip': 31.076509475708008, 'lat_1080ti_1': 0.48643498501079624, 'lat_1080ti_32': 0.34702765432760885, 'lat_1080ti_64': 0.22453146617739436, 'lat_2080ti_1': 0.44340853902961347, 'lat_2080ti_32': 0.34223439941522277, 'lat_2080ti_64': 0.24457267078002504, 'lat_essential_ph_1': 0.37735849056603776, 'lat_eyeriss': 0.3023582051235031, 'lat_fpga': 0.30234813326868454, 'lat_gold_6226': 0.3659161936197356, 'lat_gold_6240': 0.45901213224356774, 'lat_pixel2': 0.3695652173913043, 'lat_pixel3': 0.313352230595686, 'lat_raspi4': 0.3864790511178914, 'lat_samsung_a50': 0.15789473684210525, 'lat_samsung_s7': 0.1968503937007874, 'lat_silver_4114': 0.4699394776906373, 'lat_silver_4210r': 0.45513410277182564, 'lat_titan_rtx_1': 0.408654394316924, 'lat_titan_rtx_32': 0.340475890281835, 'lat_titan_rtx_64': 0.2547613755255515, 'lat_titanx_1': 0.22087404859138868, 'lat_titanx_32': 0.26950713339456545, 'lat_titanx_64': 0.20793581298040453, 'lat_titanxp_1': 0.4093129895875271, 'lat_titanxp_32': 0.30355175472919577, 'lat_titanxp_64': 0.23268108261969023}
| |
FBNet_421
|
FBNet
|
421
|
421
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_713[FLOAT, 16x3x3x3]
%onnx::Conv_714[FLOAT, 16]
%onnx::Conv_716[FLOAT, 48x16x1x1]
%onnx::Conv_717[FLOAT, 48]
%onnx::Conv_719[FLOAT, 48x1x5x5]
%onnx::Conv_722[FLOAT, 16x48x1x1]
%onnx::Conv_725[FLOAT, 96x16x1x1]
%onnx::Conv_726[FLOAT, 96]
%onnx::Conv_728[FLOAT, 96x1x5x5]
%onnx::Conv_731[FLOAT, 24x96x1x1]
%onnx::Conv_732[FLOAT, 24]
%onnx::Conv_734[FLOAT, 24x24x1x1]
%onnx::Conv_737[FLOAT, 24x1x5x5]
%onnx::Conv_740[FLOAT, 24x24x1x1]
%onnx::Conv_743[FLOAT, 144x24x1x1]
%onnx::Conv_744[FLOAT, 144]
%onnx::Conv_746[FLOAT, 144x1x5x5]
%onnx::Conv_749[FLOAT, 24x144x1x1]
%onnx::Conv_752[FLOAT, 24x12x1x1]
%onnx::Conv_755[FLOAT, 24x1x5x5]
%onnx::Conv_758[FLOAT, 24x12x1x1]
%onnx::Conv_761[FLOAT, 24x12x1x1]
%onnx::Conv_764[FLOAT, 24x1x3x3]
%onnx::Conv_767[FLOAT, 32x12x1x1]
%onnx::Conv_768[FLOAT, 32]
%onnx::Conv_770[FLOAT, 32x16x1x1]
%onnx::Conv_773[FLOAT, 32x1x5x5]
%onnx::Conv_776[FLOAT, 32x16x1x1]
%onnx::Conv_779[FLOAT, 96x32x1x1]
%onnx::Conv_782[FLOAT, 96x1x3x3]
%onnx::Conv_785[FLOAT, 32x96x1x1]
%onnx::Conv_788[FLOAT, 32x32x1x1]
%onnx::Conv_791[FLOAT, 32x1x3x3]
%onnx::Conv_794[FLOAT, 32x32x1x1]
%onnx::Conv_797[FLOAT, 96x32x1x1]
%onnx::Conv_800[FLOAT, 96x1x5x5]
%onnx::Conv_803[FLOAT, 64x96x1x1]
%onnx::Conv_804[FLOAT, 64]
%onnx::Conv_806[FLOAT, 64x32x1x1]
%onnx::Conv_809[FLOAT, 64x1x5x5]
%onnx::Conv_812[FLOAT, 64x32x1x1]
%onnx::Conv_815[FLOAT, 64x64x1x1]
%onnx::Conv_818[FLOAT, 64x1x3x3]
%onnx::Conv_821[FLOAT, 64x64x1x1]
%onnx::Conv_824[FLOAT, 192x64x1x1]
%onnx::Conv_825[FLOAT, 192]
%onnx::Conv_827[FLOAT, 192x1x5x5]
%onnx::Conv_830[FLOAT, 64x192x1x1]
%onnx::Conv_833[FLOAT, 384x64x1x1]
%onnx::Conv_834[FLOAT, 384]
%onnx::Conv_836[FLOAT, 384x1x3x3]
%onnx::Conv_839[FLOAT, 112x384x1x1]
%onnx::Conv_840[FLOAT, 112]
%onnx::Conv_842[FLOAT, 112x112x1x1]
%onnx::Conv_845[FLOAT, 112x1x5x5]
%onnx::Conv_848[FLOAT, 112x112x1x1]
%onnx::Conv_851[FLOAT, 672x112x1x1]
%onnx::Conv_852[FLOAT, 672]
%onnx::Conv_854[FLOAT, 672x1x5x5]
%onnx::Conv_857[FLOAT, 112x672x1x1]
%onnx::Conv_860[FLOAT, 112x112x1x1]
%onnx::Conv_863[FLOAT, 112x1x5x5]
%onnx::Conv_866[FLOAT, 112x112x1x1]
%onnx::Conv_869[FLOAT, 112x112x1x1]
%onnx::Conv_872[FLOAT, 112x1x5x5]
%onnx::Conv_875[FLOAT, 184x112x1x1]
%onnx::Conv_876[FLOAT, 184]
%onnx::Conv_878[FLOAT, 184x92x1x1]
%onnx::Conv_881[FLOAT, 184x1x5x5]
%onnx::Conv_884[FLOAT, 184x92x1x1]
%onnx::Conv_887[FLOAT, 552x184x1x1]
%onnx::Conv_888[FLOAT, 552]
%onnx::Conv_890[FLOAT, 552x1x3x3]
%onnx::Conv_893[FLOAT, 184x552x1x1]
%onnx::Conv_896[FLOAT, 552x184x1x1]
%onnx::Conv_899[FLOAT, 552x1x5x5]
%onnx::Conv_902[FLOAT, 184x552x1x1]
%onnx::Conv_905[FLOAT, 1104x184x1x1]
%onnx::Conv_906[FLOAT, 1104]
%onnx::Conv_908[FLOAT, 1104x1x3x3]
%onnx::Conv_911[FLOAT, 352x1104x1x1]
%onnx::Conv_912[FLOAT, 352]
%onnx::Conv_914[FLOAT, 1504x352x1x1]
%onnx::Conv_915[FLOAT, 1504]
) {
%onnx::Conv_909 = Identity(%onnx::Conv_906)
%onnx::Conv_903 = Identity(%onnx::Conv_876)
%onnx::Conv_900 = Identity(%onnx::Conv_888)
%onnx::Conv_897 = Identity(%onnx::Conv_888)
%onnx::Conv_894 = Identity(%onnx::Conv_876)
%onnx::Conv_891 = Identity(%onnx::Conv_888)
%onnx::Conv_885 = Identity(%onnx::Conv_876)
%onnx::Conv_882 = Identity(%onnx::Conv_876)
%onnx::Conv_879 = Identity(%onnx::Conv_876)
%onnx::Conv_873 = Identity(%onnx::Conv_840)
%onnx::Conv_870 = Identity(%onnx::Conv_840)
%onnx::Conv_867 = Identity(%onnx::Conv_840)
%onnx::Conv_864 = Identity(%onnx::Conv_840)
%onnx::Conv_861 = Identity(%onnx::Conv_840)
%onnx::Conv_858 = Identity(%onnx::Conv_840)
%onnx::Conv_855 = Identity(%onnx::Conv_852)
%onnx::Conv_849 = Identity(%onnx::Conv_840)
%onnx::Conv_846 = Identity(%onnx::Conv_840)
%onnx::Conv_843 = Identity(%onnx::Conv_840)
%onnx::Conv_837 = Identity(%onnx::Conv_834)
%onnx::Conv_831 = Identity(%onnx::Conv_804)
%onnx::Conv_828 = Identity(%onnx::Conv_825)
%onnx::Conv_822 = Identity(%onnx::Conv_804)
%onnx::Conv_819 = Identity(%onnx::Conv_804)
%onnx::Conv_816 = Identity(%onnx::Conv_804)
%onnx::Conv_813 = Identity(%onnx::Conv_804)
%onnx::Conv_810 = Identity(%onnx::Conv_804)
%onnx::Conv_807 = Identity(%onnx::Conv_804)
%onnx::Conv_801 = Identity(%onnx::Conv_726)
%onnx::Conv_798 = Identity(%onnx::Conv_726)
%onnx::Conv_795 = Identity(%onnx::Conv_768)
%onnx::Conv_792 = Identity(%onnx::Conv_768)
%onnx::Conv_789 = Identity(%onnx::Conv_768)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_726)
%onnx::Conv_780 = Identity(%onnx::Conv_726)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_768)
%onnx::Conv_771 = Identity(%onnx::Conv_768)
%onnx::Conv_765 = Identity(%onnx::Conv_732)
%onnx::Conv_762 = Identity(%onnx::Conv_732)
%onnx::Conv_759 = Identity(%onnx::Conv_732)
%onnx::Conv_756 = Identity(%onnx::Conv_732)
%onnx::Conv_753 = Identity(%onnx::Conv_732)
%onnx::Conv_750 = Identity(%onnx::Conv_732)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_741 = Identity(%onnx::Conv_732)
%onnx::Conv_738 = Identity(%onnx::Conv_732)
%onnx::Conv_735 = Identity(%onnx::Conv_732)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_713, %onnx::Conv_714)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_872, %onnx::Conv_873)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_875, %onnx::Conv_876)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_878, %onnx::Conv_879)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_881, %onnx::Conv_882)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_884, %onnx::Conv_885)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_887, %onnx::Conv_888)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_890, %onnx::Conv_891)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_893, %onnx::Conv_894)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_896, %onnx::Conv_897)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_899, %onnx::Conv_900)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_902, %onnx::Conv_903)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_905, %onnx::Conv_906)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_908, %onnx::Conv_909)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_911, %onnx::Conv_912)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_914, %onnx::Conv_915)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%711 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %711
}
|
val_accuracy
| 0 | 79,857,792 | 2,191,524 |
{'zcp_synflow': 84.32266695231384, 'zcp_zen': 74.4598617553711, 'zcp_epe_nas': 29.006947485420152, 'zcp_fisher': 0.20871317386627197, 'zcp_flops': 79857792.0, 'zcp_grad_norm': 30.775985717773438, 'zcp_grasp': -0.17218017578125, 'zcp_jacov': -16.054615020424617, 'zcp_l2_norm': 672.15478515625, 'zcp_nwot': 213.89530528179776, 'zcp_params': 2191524.0, 'zcp_plain': -0.002169553190469742, 'zcp_snip': 59.218116760253906, 'lat_1080ti_1': 0.7109965249458524, 'lat_1080ti_32': 0.8261482860482805, 'lat_1080ti_64': 0.7043448886400135, 'lat_2080ti_1': 0.7916125147737725, 'lat_2080ti_32': 0.7752671262931903, 'lat_2080ti_64': 0.6803896140596094, 'lat_essential_ph_1': 0.32075471698113206, 'lat_eyeriss': 0.5961161131791761, 'lat_fpga': 0.60641803303021, 'lat_gold_6226': 0.40224628711633936, 'lat_gold_6240': 0.677613919014565, 'lat_pixel2': 0.5434782608695652, 'lat_pixel3': 0.6510169456984815, 'lat_raspi4': 0.6885430621652606, 'lat_samsung_a50': 0.25263157894736843, 'lat_samsung_s7': 0.2677165354330709, 'lat_silver_4114': 0.712925982918969, 'lat_silver_4210r': 0.7564007258083442, 'lat_titan_rtx_1': 0.7702797344999902, 'lat_titan_rtx_32': 0.7699889535751928, 'lat_titan_rtx_64': 0.7336704993960871, 'lat_titanx_1': 0.42227426387515254, 'lat_titanx_32': 0.7751241367341973, 'lat_titanx_64': 0.6716853965779305, 'lat_titanxp_1': 0.8879233813764842, 'lat_titanxp_32': 0.8052258079771447, 'lat_titanxp_64': 0.7063986721703915}
| |
FBNet_4642
|
FBNet
|
4642
|
4642
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_632[FLOAT, 16x3x3x3]
%onnx::Conv_633[FLOAT, 16]
%onnx::Conv_635[FLOAT, 96x16x1x1]
%onnx::Conv_636[FLOAT, 96]
%onnx::Conv_638[FLOAT, 96x1x3x3]
%onnx::Conv_641[FLOAT, 24x96x1x1]
%onnx::Conv_642[FLOAT, 24]
%onnx::Conv_644[FLOAT, 144x24x1x1]
%onnx::Conv_645[FLOAT, 144]
%onnx::Conv_647[FLOAT, 144x1x5x5]
%onnx::Conv_650[FLOAT, 24x144x1x1]
%onnx::Conv_653[FLOAT, 144x24x1x1]
%onnx::Conv_656[FLOAT, 144x1x3x3]
%onnx::Conv_659[FLOAT, 24x144x1x1]
%onnx::Conv_662[FLOAT, 24x12x1x1]
%onnx::Conv_665[FLOAT, 24x1x3x3]
%onnx::Conv_668[FLOAT, 32x12x1x1]
%onnx::Conv_669[FLOAT, 32]
%onnx::Conv_671[FLOAT, 32x32x1x1]
%onnx::Conv_674[FLOAT, 32x1x5x5]
%onnx::Conv_677[FLOAT, 32x32x1x1]
%onnx::Conv_680[FLOAT, 32x16x1x1]
%onnx::Conv_683[FLOAT, 32x1x5x5]
%onnx::Conv_686[FLOAT, 32x16x1x1]
%onnx::Conv_689[FLOAT, 32x16x1x1]
%onnx::Conv_692[FLOAT, 32x1x5x5]
%onnx::Conv_695[FLOAT, 32x16x1x1]
%onnx::Conv_698[FLOAT, 32x32x1x1]
%onnx::Conv_701[FLOAT, 32x1x3x3]
%onnx::Conv_704[FLOAT, 64x32x1x1]
%onnx::Conv_705[FLOAT, 64]
%onnx::Conv_707[FLOAT, 384x64x1x1]
%onnx::Conv_708[FLOAT, 384]
%onnx::Conv_710[FLOAT, 384x1x3x3]
%onnx::Conv_713[FLOAT, 64x384x1x1]
%onnx::Conv_716[FLOAT, 64x64x1x1]
%onnx::Conv_719[FLOAT, 64x1x5x5]
%onnx::Conv_722[FLOAT, 64x64x1x1]
%onnx::Conv_725[FLOAT, 64x32x1x1]
%onnx::Conv_728[FLOAT, 64x1x5x5]
%onnx::Conv_731[FLOAT, 64x32x1x1]
%onnx::Conv_734[FLOAT, 384x64x1x1]
%onnx::Conv_737[FLOAT, 384x1x5x5]
%onnx::Conv_740[FLOAT, 112x384x1x1]
%onnx::Conv_741[FLOAT, 112]
%onnx::Conv_743[FLOAT, 112x56x1x1]
%onnx::Conv_746[FLOAT, 112x1x3x3]
%onnx::Conv_749[FLOAT, 112x56x1x1]
%onnx::Conv_752[FLOAT, 336x112x1x1]
%onnx::Conv_753[FLOAT, 336]
%onnx::Conv_755[FLOAT, 336x1x3x3]
%onnx::Conv_758[FLOAT, 112x336x1x1]
%onnx::Conv_761[FLOAT, 112x112x1x1]
%onnx::Conv_764[FLOAT, 112x1x3x3]
%onnx::Conv_767[FLOAT, 184x112x1x1]
%onnx::Conv_768[FLOAT, 184]
%onnx::Conv_770[FLOAT, 184x184x1x1]
%onnx::Conv_773[FLOAT, 184x1x5x5]
%onnx::Conv_776[FLOAT, 184x184x1x1]
%onnx::Conv_779[FLOAT, 184x184x1x1]
%onnx::Conv_782[FLOAT, 184x1x5x5]
%onnx::Conv_785[FLOAT, 184x184x1x1]
%onnx::Conv_788[FLOAT, 184x184x1x1]
%onnx::Conv_791[FLOAT, 184x1x3x3]
%onnx::Conv_794[FLOAT, 184x184x1x1]
%onnx::Conv_797[FLOAT, 184x184x1x1]
%onnx::Conv_800[FLOAT, 184x1x3x3]
%onnx::Conv_803[FLOAT, 352x184x1x1]
%onnx::Conv_804[FLOAT, 352]
%onnx::Conv_806[FLOAT, 1504x352x1x1]
%onnx::Conv_807[FLOAT, 1504]
) {
%onnx::Conv_801 = Identity(%onnx::Conv_768)
%onnx::Conv_798 = Identity(%onnx::Conv_768)
%onnx::Conv_795 = Identity(%onnx::Conv_768)
%onnx::Conv_792 = Identity(%onnx::Conv_768)
%onnx::Conv_789 = Identity(%onnx::Conv_768)
%onnx::Conv_786 = Identity(%onnx::Conv_768)
%onnx::Conv_783 = Identity(%onnx::Conv_768)
%onnx::Conv_780 = Identity(%onnx::Conv_768)
%onnx::Conv_777 = Identity(%onnx::Conv_768)
%onnx::Conv_774 = Identity(%onnx::Conv_768)
%onnx::Conv_771 = Identity(%onnx::Conv_768)
%onnx::Conv_765 = Identity(%onnx::Conv_741)
%onnx::Conv_762 = Identity(%onnx::Conv_741)
%onnx::Conv_759 = Identity(%onnx::Conv_741)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_750 = Identity(%onnx::Conv_741)
%onnx::Conv_747 = Identity(%onnx::Conv_741)
%onnx::Conv_744 = Identity(%onnx::Conv_741)
%onnx::Conv_738 = Identity(%onnx::Conv_708)
%onnx::Conv_735 = Identity(%onnx::Conv_708)
%onnx::Conv_732 = Identity(%onnx::Conv_705)
%onnx::Conv_729 = Identity(%onnx::Conv_705)
%onnx::Conv_726 = Identity(%onnx::Conv_705)
%onnx::Conv_723 = Identity(%onnx::Conv_705)
%onnx::Conv_720 = Identity(%onnx::Conv_705)
%onnx::Conv_717 = Identity(%onnx::Conv_705)
%onnx::Conv_714 = Identity(%onnx::Conv_705)
%onnx::Conv_711 = Identity(%onnx::Conv_708)
%onnx::Conv_702 = Identity(%onnx::Conv_669)
%onnx::Conv_699 = Identity(%onnx::Conv_669)
%onnx::Conv_696 = Identity(%onnx::Conv_669)
%onnx::Conv_693 = Identity(%onnx::Conv_669)
%onnx::Conv_690 = Identity(%onnx::Conv_669)
%onnx::Conv_687 = Identity(%onnx::Conv_669)
%onnx::Conv_684 = Identity(%onnx::Conv_669)
%onnx::Conv_681 = Identity(%onnx::Conv_669)
%onnx::Conv_678 = Identity(%onnx::Conv_669)
%onnx::Conv_675 = Identity(%onnx::Conv_669)
%onnx::Conv_672 = Identity(%onnx::Conv_669)
%onnx::Conv_666 = Identity(%onnx::Conv_642)
%onnx::Conv_663 = Identity(%onnx::Conv_642)
%onnx::Conv_660 = Identity(%onnx::Conv_642)
%onnx::Conv_657 = Identity(%onnx::Conv_645)
%onnx::Conv_654 = Identity(%onnx::Conv_645)
%onnx::Conv_651 = Identity(%onnx::Conv_642)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_632, %onnx::Conv_633)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%630 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %630
}
|
val_accuracy
| 0 | 60,082,048 | 1,317,772 |
{'zcp_synflow': 72.13269911652658, 'zcp_zen': 63.37042999267578, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.07011188566684723, 'zcp_flops': 60082048.0, 'zcp_grad_norm': 19.626596450805664, 'zcp_grasp': -0.07778358459472656, 'zcp_jacov': -16.05069235554498, 'zcp_l2_norm': 550.0938720703125, 'zcp_nwot': 213.24698967567497, 'zcp_params': 1317772.0, 'zcp_plain': 0.0016791797243058681, 'zcp_snip': 38.06217575073242, 'lat_1080ti_1': 0.48094520228692733, 'lat_1080ti_32': 0.6706306981415354, 'lat_1080ti_64': 0.5405687338957788, 'lat_2080ti_1': 0.5525856547825676, 'lat_2080ti_32': 0.6336944708176282, 'lat_2080ti_64': 0.5718755740054629, 'lat_essential_ph_1': 0.1320754716981132, 'lat_eyeriss': 0.36620028239781405, 'lat_fpga': 0.35737539457148787, 'lat_gold_6226': 0.14470185821309728, 'lat_gold_6240': 0.2541468874175946, 'lat_pixel2': 0.17391304347826086, 'lat_pixel3': 0.37432222520523334, 'lat_raspi4': 0.36531501467551375, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.06299212598425197, 'lat_silver_4114': 0.2698451515941344, 'lat_silver_4210r': 0.2990363894847731, 'lat_titan_rtx_1': 0.4908775307270161, 'lat_titan_rtx_32': 0.5826634990193614, 'lat_titan_rtx_64': 0.5862448418332404, 'lat_titanx_1': 0.26070461672605094, 'lat_titanx_32': 0.5967924846386908, 'lat_titanx_64': 0.5112998024245528, 'lat_titanxp_1': 0.4567841292807143, 'lat_titanxp_32': 0.5853338099108395, 'lat_titanxp_64': 0.55494055045086}
| |
FBNet_4563
|
FBNet
|
4563
|
4563
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_677[FLOAT, 16x3x3x3]
%onnx::Conv_678[FLOAT, 16]
%onnx::Conv_680[FLOAT, 16x16x1x1]
%onnx::Conv_683[FLOAT, 16x1x3x3]
%onnx::Conv_686[FLOAT, 16x16x1x1]
%onnx::Conv_689[FLOAT, 48x16x1x1]
%onnx::Conv_690[FLOAT, 48]
%onnx::Conv_692[FLOAT, 48x1x5x5]
%onnx::Conv_695[FLOAT, 24x48x1x1]
%onnx::Conv_696[FLOAT, 24]
%onnx::Conv_698[FLOAT, 24x24x1x1]
%onnx::Conv_701[FLOAT, 24x1x3x3]
%onnx::Conv_704[FLOAT, 24x24x1x1]
%onnx::Conv_707[FLOAT, 144x24x1x1]
%onnx::Conv_708[FLOAT, 144]
%onnx::Conv_710[FLOAT, 144x1x5x5]
%onnx::Conv_713[FLOAT, 24x144x1x1]
%onnx::Conv_716[FLOAT, 144x24x1x1]
%onnx::Conv_719[FLOAT, 144x1x5x5]
%onnx::Conv_722[FLOAT, 32x144x1x1]
%onnx::Conv_723[FLOAT, 32]
%onnx::Conv_725[FLOAT, 96x32x1x1]
%onnx::Conv_726[FLOAT, 96]
%onnx::Conv_728[FLOAT, 96x1x5x5]
%onnx::Conv_731[FLOAT, 32x96x1x1]
%onnx::Conv_734[FLOAT, 32x16x1x1]
%onnx::Conv_737[FLOAT, 32x1x3x3]
%onnx::Conv_740[FLOAT, 32x16x1x1]
%onnx::Conv_743[FLOAT, 32x16x1x1]
%onnx::Conv_746[FLOAT, 32x1x3x3]
%onnx::Conv_749[FLOAT, 32x16x1x1]
%onnx::Conv_752[FLOAT, 32x16x1x1]
%onnx::Conv_755[FLOAT, 32x1x3x3]
%onnx::Conv_758[FLOAT, 64x16x1x1]
%onnx::Conv_759[FLOAT, 64]
%onnx::Conv_761[FLOAT, 192x64x1x1]
%onnx::Conv_762[FLOAT, 192]
%onnx::Conv_764[FLOAT, 192x1x5x5]
%onnx::Conv_767[FLOAT, 64x192x1x1]
%onnx::Conv_770[FLOAT, 192x64x1x1]
%onnx::Conv_773[FLOAT, 192x1x5x5]
%onnx::Conv_776[FLOAT, 64x192x1x1]
%onnx::Conv_779[FLOAT, 64x64x1x1]
%onnx::Conv_782[FLOAT, 64x1x3x3]
%onnx::Conv_785[FLOAT, 64x64x1x1]
%onnx::Conv_788[FLOAT, 64x32x1x1]
%onnx::Conv_791[FLOAT, 64x1x5x5]
%onnx::Conv_794[FLOAT, 112x32x1x1]
%onnx::Conv_795[FLOAT, 112]
%onnx::Conv_797[FLOAT, 336x112x1x1]
%onnx::Conv_798[FLOAT, 336]
%onnx::Conv_800[FLOAT, 336x1x3x3]
%onnx::Conv_803[FLOAT, 112x336x1x1]
%onnx::Conv_806[FLOAT, 112x56x1x1]
%onnx::Conv_809[FLOAT, 112x1x5x5]
%onnx::Conv_812[FLOAT, 112x56x1x1]
%onnx::Conv_815[FLOAT, 336x112x1x1]
%onnx::Conv_818[FLOAT, 336x1x5x5]
%onnx::Conv_821[FLOAT, 184x336x1x1]
%onnx::Conv_822[FLOAT, 184]
%onnx::Conv_824[FLOAT, 184x184x1x1]
%onnx::Conv_827[FLOAT, 184x1x5x5]
%onnx::Conv_830[FLOAT, 184x184x1x1]
%onnx::Conv_833[FLOAT, 184x184x1x1]
%onnx::Conv_836[FLOAT, 184x1x5x5]
%onnx::Conv_839[FLOAT, 184x184x1x1]
%onnx::Conv_842[FLOAT, 552x184x1x1]
%onnx::Conv_843[FLOAT, 552]
%onnx::Conv_845[FLOAT, 552x1x3x3]
%onnx::Conv_848[FLOAT, 184x552x1x1]
%onnx::Conv_851[FLOAT, 184x92x1x1]
%onnx::Conv_854[FLOAT, 184x1x5x5]
%onnx::Conv_857[FLOAT, 352x92x1x1]
%onnx::Conv_858[FLOAT, 352]
%onnx::Conv_860[FLOAT, 1504x352x1x1]
%onnx::Conv_861[FLOAT, 1504]
) {
%onnx::Conv_855 = Identity(%onnx::Conv_822)
%onnx::Conv_852 = Identity(%onnx::Conv_822)
%onnx::Conv_849 = Identity(%onnx::Conv_822)
%onnx::Conv_846 = Identity(%onnx::Conv_843)
%onnx::Conv_840 = Identity(%onnx::Conv_822)
%onnx::Conv_837 = Identity(%onnx::Conv_822)
%onnx::Conv_834 = Identity(%onnx::Conv_822)
%onnx::Conv_831 = Identity(%onnx::Conv_822)
%onnx::Conv_828 = Identity(%onnx::Conv_822)
%onnx::Conv_825 = Identity(%onnx::Conv_822)
%onnx::Conv_819 = Identity(%onnx::Conv_798)
%onnx::Conv_816 = Identity(%onnx::Conv_798)
%onnx::Conv_813 = Identity(%onnx::Conv_795)
%onnx::Conv_810 = Identity(%onnx::Conv_795)
%onnx::Conv_807 = Identity(%onnx::Conv_795)
%onnx::Conv_804 = Identity(%onnx::Conv_795)
%onnx::Conv_801 = Identity(%onnx::Conv_798)
%onnx::Conv_792 = Identity(%onnx::Conv_759)
%onnx::Conv_789 = Identity(%onnx::Conv_759)
%onnx::Conv_786 = Identity(%onnx::Conv_759)
%onnx::Conv_783 = Identity(%onnx::Conv_759)
%onnx::Conv_780 = Identity(%onnx::Conv_759)
%onnx::Conv_777 = Identity(%onnx::Conv_759)
%onnx::Conv_774 = Identity(%onnx::Conv_762)
%onnx::Conv_771 = Identity(%onnx::Conv_762)
%onnx::Conv_768 = Identity(%onnx::Conv_759)
%onnx::Conv_765 = Identity(%onnx::Conv_762)
%onnx::Conv_756 = Identity(%onnx::Conv_723)
%onnx::Conv_753 = Identity(%onnx::Conv_723)
%onnx::Conv_750 = Identity(%onnx::Conv_723)
%onnx::Conv_747 = Identity(%onnx::Conv_723)
%onnx::Conv_744 = Identity(%onnx::Conv_723)
%onnx::Conv_741 = Identity(%onnx::Conv_723)
%onnx::Conv_738 = Identity(%onnx::Conv_723)
%onnx::Conv_735 = Identity(%onnx::Conv_723)
%onnx::Conv_732 = Identity(%onnx::Conv_723)
%onnx::Conv_729 = Identity(%onnx::Conv_726)
%onnx::Conv_720 = Identity(%onnx::Conv_708)
%onnx::Conv_717 = Identity(%onnx::Conv_708)
%onnx::Conv_714 = Identity(%onnx::Conv_696)
%onnx::Conv_711 = Identity(%onnx::Conv_708)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_696)
%onnx::Conv_699 = Identity(%onnx::Conv_696)
%onnx::Conv_693 = Identity(%onnx::Conv_690)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_678)
%onnx::Conv_681 = Identity(%onnx::Conv_678)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_677, %onnx::Conv_678)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%675 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %675
}
|
val_accuracy
| 0 | 57,401,216 | 1,422,236 |
{'zcp_synflow': 74.85088981112386, 'zcp_zen': 65.21505737304688, 'zcp_epe_nas': 21.753737046960442, 'zcp_fisher': 0.08163570612668991, 'zcp_flops': 57401216.0, 'zcp_grad_norm': 24.044172286987305, 'zcp_grasp': -0.0042209625244140625, 'zcp_jacov': -16.056465035558475, 'zcp_l2_norm': 563.458984375, 'zcp_nwot': 211.30088965259358, 'zcp_params': 1422236.0, 'zcp_plain': -0.0014816473703831434, 'zcp_snip': 41.31752014160156, 'lat_1080ti_1': 0.5863427483949643, 'lat_1080ti_32': 0.6242747119444967, 'lat_1080ti_64': 0.48813086583038473, 'lat_2080ti_1': 0.6565310606860907, 'lat_2080ti_32': 0.6258876608628408, 'lat_2080ti_64': 0.5253450146095762, 'lat_essential_ph_1': 0.22641509433962265, 'lat_eyeriss': 0.320191443712981, 'lat_fpga': 0.2559890756884592, 'lat_gold_6226': 0.2063777671383377, 'lat_gold_6240': 0.30395821899571707, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.3562641404008034, 'lat_raspi4': 0.36029922120964497, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.11023622047244094, 'lat_silver_4114': 0.33691157984997805, 'lat_silver_4210r': 0.3381797562410162, 'lat_titan_rtx_1': 0.6197940479492162, 'lat_titan_rtx_32': 0.6078063872006011, 'lat_titan_rtx_64': 0.5600340752531945, 'lat_titanx_1': 0.3163635801713144, 'lat_titanx_32': 0.5832279730600535, 'lat_titanx_64': 0.46985048401553975, 'lat_titanxp_1': 0.5676034854833184, 'lat_titanxp_32': 0.6049545488598032, 'lat_titanxp_64': 0.534275249726476}
| |
FBNet_4439
|
FBNet
|
4439
|
4439
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_652[FLOAT, 16x3x3x3]
%onnx::Conv_653[FLOAT, 16]
%onnx::Conv_655[FLOAT, 16x8x1x1]
%onnx::Conv_658[FLOAT, 16x1x5x5]
%onnx::Conv_661[FLOAT, 16x8x1x1]
%onnx::Conv_664[FLOAT, 96x16x1x1]
%onnx::Conv_665[FLOAT, 96]
%onnx::Conv_667[FLOAT, 96x1x5x5]
%onnx::Conv_670[FLOAT, 24x96x1x1]
%onnx::Conv_671[FLOAT, 24]
%onnx::Conv_673[FLOAT, 24x12x1x1]
%onnx::Conv_676[FLOAT, 24x1x5x5]
%onnx::Conv_679[FLOAT, 24x12x1x1]
%onnx::Conv_682[FLOAT, 24x24x1x1]
%onnx::Conv_685[FLOAT, 24x1x3x3]
%onnx::Conv_688[FLOAT, 24x24x1x1]
%onnx::Conv_691[FLOAT, 24x24x1x1]
%onnx::Conv_694[FLOAT, 24x1x3x3]
%onnx::Conv_697[FLOAT, 24x24x1x1]
%onnx::Conv_700[FLOAT, 144x24x1x1]
%onnx::Conv_701[FLOAT, 144]
%onnx::Conv_703[FLOAT, 144x1x5x5]
%onnx::Conv_706[FLOAT, 32x144x1x1]
%onnx::Conv_707[FLOAT, 32]
%onnx::Conv_709[FLOAT, 32x32x1x1]
%onnx::Conv_712[FLOAT, 32x1x5x5]
%onnx::Conv_715[FLOAT, 32x32x1x1]
%onnx::Conv_718[FLOAT, 192x32x1x1]
%onnx::Conv_719[FLOAT, 192]
%onnx::Conv_721[FLOAT, 192x1x3x3]
%onnx::Conv_724[FLOAT, 32x192x1x1]
%onnx::Conv_727[FLOAT, 96x32x1x1]
%onnx::Conv_730[FLOAT, 96x1x5x5]
%onnx::Conv_733[FLOAT, 64x96x1x1]
%onnx::Conv_734[FLOAT, 64]
%onnx::Conv_736[FLOAT, 64x32x1x1]
%onnx::Conv_739[FLOAT, 64x1x3x3]
%onnx::Conv_742[FLOAT, 64x32x1x1]
%onnx::Conv_745[FLOAT, 64x32x1x1]
%onnx::Conv_748[FLOAT, 64x1x5x5]
%onnx::Conv_751[FLOAT, 64x32x1x1]
%onnx::Conv_754[FLOAT, 112x64x1x1]
%onnx::Conv_755[FLOAT, 112]
%onnx::Conv_757[FLOAT, 336x112x1x1]
%onnx::Conv_758[FLOAT, 336]
%onnx::Conv_760[FLOAT, 336x1x3x3]
%onnx::Conv_763[FLOAT, 112x336x1x1]
%onnx::Conv_766[FLOAT, 112x112x1x1]
%onnx::Conv_769[FLOAT, 112x1x5x5]
%onnx::Conv_772[FLOAT, 112x112x1x1]
%onnx::Conv_775[FLOAT, 112x56x1x1]
%onnx::Conv_778[FLOAT, 112x1x5x5]
%onnx::Conv_781[FLOAT, 184x56x1x1]
%onnx::Conv_782[FLOAT, 184]
%onnx::Conv_784[FLOAT, 184x92x1x1]
%onnx::Conv_787[FLOAT, 184x1x5x5]
%onnx::Conv_790[FLOAT, 184x92x1x1]
%onnx::Conv_793[FLOAT, 184x92x1x1]
%onnx::Conv_796[FLOAT, 184x1x3x3]
%onnx::Conv_799[FLOAT, 184x92x1x1]
%onnx::Conv_802[FLOAT, 184x184x1x1]
%onnx::Conv_805[FLOAT, 184x1x5x5]
%onnx::Conv_808[FLOAT, 184x184x1x1]
%onnx::Conv_811[FLOAT, 1104x184x1x1]
%onnx::Conv_812[FLOAT, 1104]
%onnx::Conv_814[FLOAT, 1104x1x3x3]
%onnx::Conv_817[FLOAT, 352x1104x1x1]
%onnx::Conv_818[FLOAT, 352]
%onnx::Conv_820[FLOAT, 1504x352x1x1]
%onnx::Conv_821[FLOAT, 1504]
) {
%onnx::Conv_815 = Identity(%onnx::Conv_812)
%onnx::Conv_809 = Identity(%onnx::Conv_782)
%onnx::Conv_806 = Identity(%onnx::Conv_782)
%onnx::Conv_803 = Identity(%onnx::Conv_782)
%onnx::Conv_800 = Identity(%onnx::Conv_782)
%onnx::Conv_797 = Identity(%onnx::Conv_782)
%onnx::Conv_794 = Identity(%onnx::Conv_782)
%onnx::Conv_791 = Identity(%onnx::Conv_782)
%onnx::Conv_788 = Identity(%onnx::Conv_782)
%onnx::Conv_785 = Identity(%onnx::Conv_782)
%onnx::Conv_779 = Identity(%onnx::Conv_755)
%onnx::Conv_776 = Identity(%onnx::Conv_755)
%onnx::Conv_773 = Identity(%onnx::Conv_755)
%onnx::Conv_770 = Identity(%onnx::Conv_755)
%onnx::Conv_767 = Identity(%onnx::Conv_755)
%onnx::Conv_764 = Identity(%onnx::Conv_755)
%onnx::Conv_761 = Identity(%onnx::Conv_758)
%onnx::Conv_752 = Identity(%onnx::Conv_734)
%onnx::Conv_749 = Identity(%onnx::Conv_734)
%onnx::Conv_746 = Identity(%onnx::Conv_734)
%onnx::Conv_743 = Identity(%onnx::Conv_734)
%onnx::Conv_740 = Identity(%onnx::Conv_734)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_731 = Identity(%onnx::Conv_665)
%onnx::Conv_728 = Identity(%onnx::Conv_665)
%onnx::Conv_725 = Identity(%onnx::Conv_707)
%onnx::Conv_722 = Identity(%onnx::Conv_719)
%onnx::Conv_716 = Identity(%onnx::Conv_707)
%onnx::Conv_713 = Identity(%onnx::Conv_707)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_701)
%onnx::Conv_698 = Identity(%onnx::Conv_671)
%onnx::Conv_695 = Identity(%onnx::Conv_671)
%onnx::Conv_692 = Identity(%onnx::Conv_671)
%onnx::Conv_689 = Identity(%onnx::Conv_671)
%onnx::Conv_686 = Identity(%onnx::Conv_671)
%onnx::Conv_683 = Identity(%onnx::Conv_671)
%onnx::Conv_680 = Identity(%onnx::Conv_671)
%onnx::Conv_677 = Identity(%onnx::Conv_671)
%onnx::Conv_674 = Identity(%onnx::Conv_671)
%onnx::Conv_668 = Identity(%onnx::Conv_665)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_652, %onnx::Conv_653)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%650 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %650
}
|
val_accuracy
| 0 | 54,489,216 | 1,640,996 |
{'zcp_synflow': 68.61329683801868, 'zcp_zen': 58.71949768066406, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.07984098792076111, 'zcp_flops': 54489216.0, 'zcp_grad_norm': 20.324878692626953, 'zcp_grasp': -0.02327728271484375, 'zcp_jacov': -16.055671959133186, 'zcp_l2_norm': 505.4128112792969, 'zcp_nwot': 209.72526505771017, 'zcp_params': 1640996.0, 'zcp_plain': 0.0066939727403223515, 'zcp_snip': 29.807086944580078, 'lat_1080ti_1': 0.5669521792074429, 'lat_1080ti_32': 0.4090481305544017, 'lat_1080ti_64': 0.3711938681449474, 'lat_2080ti_1': 0.5243017901334107, 'lat_2080ti_32': 0.4595368061628108, 'lat_2080ti_64': 0.38695245702720477, 'lat_essential_ph_1': 0.1320754716981132, 'lat_eyeriss': 0.27487026204316645, 'lat_fpga': 0.2593831326254959, 'lat_gold_6226': 0.1768814823757332, 'lat_gold_6240': 0.336251571798507, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.30698488365136084, 'lat_raspi4': 0.3848590732985905, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.05511811023622047, 'lat_silver_4114': 0.37121516453118664, 'lat_silver_4210r': 0.41109188955451476, 'lat_titan_rtx_1': 0.4826491488341846, 'lat_titan_rtx_32': 0.4439450208277762, 'lat_titan_rtx_64': 0.39209565075993663, 'lat_titanx_1': 0.24872300389160412, 'lat_titanx_32': 0.4027575341263109, 'lat_titanx_64': 0.35736545695214816, 'lat_titanxp_1': 0.4418884518724316, 'lat_titanxp_32': 0.43264402404952557, 'lat_titanxp_64': 0.3731256932020014}
| |
FBNet_1604
|
FBNet
|
1604
|
1604
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_695[FLOAT, 16x3x3x3]
%onnx::Conv_696[FLOAT, 16]
%onnx::Conv_698[FLOAT, 48x16x1x1]
%onnx::Conv_699[FLOAT, 48]
%onnx::Conv_701[FLOAT, 48x1x5x5]
%onnx::Conv_704[FLOAT, 16x48x1x1]
%onnx::Conv_707[FLOAT, 48x16x1x1]
%onnx::Conv_710[FLOAT, 48x1x5x5]
%onnx::Conv_713[FLOAT, 24x48x1x1]
%onnx::Conv_714[FLOAT, 24]
%onnx::Conv_716[FLOAT, 72x24x1x1]
%onnx::Conv_717[FLOAT, 72]
%onnx::Conv_719[FLOAT, 72x1x3x3]
%onnx::Conv_722[FLOAT, 24x72x1x1]
%onnx::Conv_725[FLOAT, 72x24x1x1]
%onnx::Conv_728[FLOAT, 72x1x3x3]
%onnx::Conv_731[FLOAT, 24x72x1x1]
%onnx::Conv_734[FLOAT, 144x24x1x1]
%onnx::Conv_735[FLOAT, 144]
%onnx::Conv_737[FLOAT, 144x1x5x5]
%onnx::Conv_740[FLOAT, 24x144x1x1]
%onnx::Conv_743[FLOAT, 24x12x1x1]
%onnx::Conv_746[FLOAT, 24x1x5x5]
%onnx::Conv_749[FLOAT, 32x12x1x1]
%onnx::Conv_750[FLOAT, 32]
%onnx::Conv_752[FLOAT, 192x32x1x1]
%onnx::Conv_753[FLOAT, 192]
%onnx::Conv_755[FLOAT, 192x1x3x3]
%onnx::Conv_758[FLOAT, 32x192x1x1]
%onnx::Conv_761[FLOAT, 32x16x1x1]
%onnx::Conv_764[FLOAT, 32x1x5x5]
%onnx::Conv_767[FLOAT, 32x16x1x1]
%onnx::Conv_770[FLOAT, 32x32x1x1]
%onnx::Conv_773[FLOAT, 32x1x5x5]
%onnx::Conv_776[FLOAT, 32x32x1x1]
%onnx::Conv_779[FLOAT, 192x32x1x1]
%onnx::Conv_782[FLOAT, 192x1x5x5]
%onnx::Conv_785[FLOAT, 64x192x1x1]
%onnx::Conv_786[FLOAT, 64]
%onnx::Conv_788[FLOAT, 64x32x1x1]
%onnx::Conv_791[FLOAT, 64x1x3x3]
%onnx::Conv_794[FLOAT, 64x32x1x1]
%onnx::Conv_797[FLOAT, 64x64x1x1]
%onnx::Conv_800[FLOAT, 64x1x3x3]
%onnx::Conv_803[FLOAT, 64x64x1x1]
%onnx::Conv_806[FLOAT, 192x64x1x1]
%onnx::Conv_809[FLOAT, 192x1x3x3]
%onnx::Conv_812[FLOAT, 64x192x1x1]
%onnx::Conv_815[FLOAT, 192x64x1x1]
%onnx::Conv_818[FLOAT, 192x1x3x3]
%onnx::Conv_821[FLOAT, 112x192x1x1]
%onnx::Conv_822[FLOAT, 112]
%onnx::Conv_824[FLOAT, 672x112x1x1]
%onnx::Conv_825[FLOAT, 672]
%onnx::Conv_827[FLOAT, 672x1x5x5]
%onnx::Conv_830[FLOAT, 112x672x1x1]
%onnx::Conv_833[FLOAT, 112x56x1x1]
%onnx::Conv_836[FLOAT, 112x1x5x5]
%onnx::Conv_839[FLOAT, 112x56x1x1]
%onnx::Conv_842[FLOAT, 112x112x1x1]
%onnx::Conv_845[FLOAT, 112x1x5x5]
%onnx::Conv_848[FLOAT, 112x112x1x1]
%onnx::Conv_851[FLOAT, 672x112x1x1]
%onnx::Conv_854[FLOAT, 672x1x5x5]
%onnx::Conv_857[FLOAT, 184x672x1x1]
%onnx::Conv_858[FLOAT, 184]
%onnx::Conv_860[FLOAT, 552x184x1x1]
%onnx::Conv_861[FLOAT, 552]
%onnx::Conv_863[FLOAT, 552x1x5x5]
%onnx::Conv_866[FLOAT, 184x552x1x1]
%onnx::Conv_869[FLOAT, 552x184x1x1]
%onnx::Conv_872[FLOAT, 552x1x3x3]
%onnx::Conv_875[FLOAT, 184x552x1x1]
%onnx::Conv_878[FLOAT, 1104x184x1x1]
%onnx::Conv_879[FLOAT, 1104]
%onnx::Conv_881[FLOAT, 1104x1x3x3]
%onnx::Conv_884[FLOAT, 184x1104x1x1]
%onnx::Conv_887[FLOAT, 184x184x1x1]
%onnx::Conv_890[FLOAT, 184x1x5x5]
%onnx::Conv_893[FLOAT, 352x184x1x1]
%onnx::Conv_894[FLOAT, 352]
%onnx::Conv_896[FLOAT, 1504x352x1x1]
%onnx::Conv_897[FLOAT, 1504]
) {
%onnx::Conv_891 = Identity(%onnx::Conv_858)
%onnx::Conv_888 = Identity(%onnx::Conv_858)
%onnx::Conv_885 = Identity(%onnx::Conv_858)
%onnx::Conv_882 = Identity(%onnx::Conv_879)
%onnx::Conv_876 = Identity(%onnx::Conv_858)
%onnx::Conv_873 = Identity(%onnx::Conv_861)
%onnx::Conv_870 = Identity(%onnx::Conv_861)
%onnx::Conv_867 = Identity(%onnx::Conv_858)
%onnx::Conv_864 = Identity(%onnx::Conv_861)
%onnx::Conv_855 = Identity(%onnx::Conv_825)
%onnx::Conv_852 = Identity(%onnx::Conv_825)
%onnx::Conv_849 = Identity(%onnx::Conv_822)
%onnx::Conv_846 = Identity(%onnx::Conv_822)
%onnx::Conv_843 = Identity(%onnx::Conv_822)
%onnx::Conv_840 = Identity(%onnx::Conv_822)
%onnx::Conv_837 = Identity(%onnx::Conv_822)
%onnx::Conv_834 = Identity(%onnx::Conv_822)
%onnx::Conv_831 = Identity(%onnx::Conv_822)
%onnx::Conv_828 = Identity(%onnx::Conv_825)
%onnx::Conv_819 = Identity(%onnx::Conv_753)
%onnx::Conv_816 = Identity(%onnx::Conv_753)
%onnx::Conv_813 = Identity(%onnx::Conv_786)
%onnx::Conv_810 = Identity(%onnx::Conv_753)
%onnx::Conv_807 = Identity(%onnx::Conv_753)
%onnx::Conv_804 = Identity(%onnx::Conv_786)
%onnx::Conv_801 = Identity(%onnx::Conv_786)
%onnx::Conv_798 = Identity(%onnx::Conv_786)
%onnx::Conv_795 = Identity(%onnx::Conv_786)
%onnx::Conv_792 = Identity(%onnx::Conv_786)
%onnx::Conv_789 = Identity(%onnx::Conv_786)
%onnx::Conv_783 = Identity(%onnx::Conv_753)
%onnx::Conv_780 = Identity(%onnx::Conv_753)
%onnx::Conv_777 = Identity(%onnx::Conv_750)
%onnx::Conv_774 = Identity(%onnx::Conv_750)
%onnx::Conv_771 = Identity(%onnx::Conv_750)
%onnx::Conv_768 = Identity(%onnx::Conv_750)
%onnx::Conv_765 = Identity(%onnx::Conv_750)
%onnx::Conv_762 = Identity(%onnx::Conv_750)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_747 = Identity(%onnx::Conv_714)
%onnx::Conv_744 = Identity(%onnx::Conv_714)
%onnx::Conv_741 = Identity(%onnx::Conv_714)
%onnx::Conv_738 = Identity(%onnx::Conv_735)
%onnx::Conv_732 = Identity(%onnx::Conv_714)
%onnx::Conv_729 = Identity(%onnx::Conv_717)
%onnx::Conv_726 = Identity(%onnx::Conv_717)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%onnx::Conv_711 = Identity(%onnx::Conv_699)
%onnx::Conv_708 = Identity(%onnx::Conv_699)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_695, %onnx::Conv_696)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_872, %onnx::Conv_873)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_875, %onnx::Conv_876)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_878, %onnx::Conv_879)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_881, %onnx::Conv_882)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_884, %onnx::Conv_885)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_887, %onnx::Conv_888)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_890, %onnx::Conv_891)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_893, %onnx::Conv_894)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_896, %onnx::Conv_897)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%693 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %693
}
|
val_accuracy
| 0 | 86,516,864 | 2,222,532 |
{'zcp_synflow': 85.21211018979055, 'zcp_zen': 75.83748626708984, 'zcp_epe_nas': 11.278773689957875, 'zcp_fisher': 0.23824511468410492, 'zcp_flops': 86516864.0, 'zcp_grad_norm': 35.12641143798828, 'zcp_grasp': -0.044322967529296875, 'zcp_jacov': -16.047501514649987, 'zcp_l2_norm': 709.0864868164062, 'zcp_nwot': 216.3642511071184, 'zcp_params': 2222532.0, 'zcp_plain': 0.004126658663153648, 'zcp_snip': 59.97085189819336, 'lat_1080ti_1': 0.7890827974083091, 'lat_1080ti_32': 0.7035231425349117, 'lat_1080ti_64': 0.703753018783555, 'lat_2080ti_1': 0.7812576049273655, 'lat_2080ti_32': 0.7430629236508468, 'lat_2080ti_64': 0.7198116462110856, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.6751324885848934, 'lat_fpga': 0.6613364470259946, 'lat_gold_6226': 0.483426158770064, 'lat_gold_6240': 0.6830981446496229, 'lat_pixel2': 0.45652173913043476, 'lat_pixel3': 0.6662733889828827, 'lat_raspi4': 0.6407083371520579, 'lat_samsung_a50': 0.2736842105263158, 'lat_samsung_s7': 0.23622047244094488, 'lat_silver_4114': 0.705178686045616, 'lat_silver_4210r': 0.753224184005901, 'lat_titan_rtx_1': 0.7615242322760083, 'lat_titan_rtx_32': 0.7295845870958113, 'lat_titan_rtx_64': 0.7462559201142326, 'lat_titanx_1': 0.4077604764089935, 'lat_titanx_32': 0.7388866003222435, 'lat_titanx_64': 0.7469922778103155, 'lat_titanxp_1': 0.7231546253853139, 'lat_titanxp_32': 0.7733976460627878, 'lat_titanxp_64': 0.7201157898176184}
| |
FBNet_4072
|
FBNet
|
4072
|
4072
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_769[FLOAT, 16x3x3x3]
%onnx::Conv_770[FLOAT, 16]
%onnx::Conv_772[FLOAT, 48x16x1x1]
%onnx::Conv_773[FLOAT, 48]
%onnx::Conv_775[FLOAT, 48x1x5x5]
%onnx::Conv_778[FLOAT, 16x48x1x1]
%onnx::Conv_781[FLOAT, 48x16x1x1]
%onnx::Conv_784[FLOAT, 48x1x5x5]
%onnx::Conv_787[FLOAT, 24x48x1x1]
%onnx::Conv_788[FLOAT, 24]
%onnx::Conv_790[FLOAT, 144x24x1x1]
%onnx::Conv_791[FLOAT, 144]
%onnx::Conv_793[FLOAT, 144x1x5x5]
%onnx::Conv_796[FLOAT, 24x144x1x1]
%onnx::Conv_799[FLOAT, 24x12x1x1]
%onnx::Conv_802[FLOAT, 24x1x3x3]
%onnx::Conv_805[FLOAT, 24x12x1x1]
%onnx::Conv_808[FLOAT, 144x24x1x1]
%onnx::Conv_811[FLOAT, 144x1x3x3]
%onnx::Conv_814[FLOAT, 24x144x1x1]
%onnx::Conv_817[FLOAT, 24x24x1x1]
%onnx::Conv_820[FLOAT, 24x1x5x5]
%onnx::Conv_823[FLOAT, 32x24x1x1]
%onnx::Conv_824[FLOAT, 32]
%onnx::Conv_826[FLOAT, 32x32x1x1]
%onnx::Conv_829[FLOAT, 32x1x3x3]
%onnx::Conv_832[FLOAT, 32x32x1x1]
%onnx::Conv_835[FLOAT, 32x32x1x1]
%onnx::Conv_838[FLOAT, 32x1x5x5]
%onnx::Conv_841[FLOAT, 32x32x1x1]
%onnx::Conv_844[FLOAT, 32x16x1x1]
%onnx::Conv_847[FLOAT, 32x1x3x3]
%onnx::Conv_850[FLOAT, 32x16x1x1]
%onnx::Conv_853[FLOAT, 32x16x1x1]
%onnx::Conv_856[FLOAT, 32x1x3x3]
%onnx::Conv_859[FLOAT, 64x16x1x1]
%onnx::Conv_860[FLOAT, 64]
%onnx::Conv_862[FLOAT, 384x64x1x1]
%onnx::Conv_863[FLOAT, 384]
%onnx::Conv_865[FLOAT, 384x1x5x5]
%onnx::Conv_868[FLOAT, 64x384x1x1]
%onnx::Conv_871[FLOAT, 64x32x1x1]
%onnx::Conv_874[FLOAT, 64x1x5x5]
%onnx::Conv_877[FLOAT, 64x32x1x1]
%onnx::Conv_880[FLOAT, 64x32x1x1]
%onnx::Conv_883[FLOAT, 64x1x3x3]
%onnx::Conv_886[FLOAT, 64x32x1x1]
%onnx::Conv_889[FLOAT, 64x32x1x1]
%onnx::Conv_892[FLOAT, 64x1x3x3]
%onnx::Conv_895[FLOAT, 112x32x1x1]
%onnx::Conv_896[FLOAT, 112]
%onnx::Conv_898[FLOAT, 112x56x1x1]
%onnx::Conv_901[FLOAT, 112x1x3x3]
%onnx::Conv_904[FLOAT, 112x56x1x1]
%onnx::Conv_907[FLOAT, 672x112x1x1]
%onnx::Conv_908[FLOAT, 672]
%onnx::Conv_910[FLOAT, 672x1x5x5]
%onnx::Conv_913[FLOAT, 112x672x1x1]
%onnx::Conv_916[FLOAT, 672x112x1x1]
%onnx::Conv_919[FLOAT, 672x1x5x5]
%onnx::Conv_922[FLOAT, 112x672x1x1]
%onnx::Conv_925[FLOAT, 184x112x1x1]
%onnx::Conv_926[FLOAT, 184]
%onnx::Conv_928[FLOAT, 184x92x1x1]
%onnx::Conv_931[FLOAT, 184x1x3x3]
%onnx::Conv_934[FLOAT, 184x92x1x1]
%onnx::Conv_937[FLOAT, 184x92x1x1]
%onnx::Conv_940[FLOAT, 184x1x5x5]
%onnx::Conv_943[FLOAT, 184x92x1x1]
%onnx::Conv_946[FLOAT, 552x184x1x1]
%onnx::Conv_947[FLOAT, 552]
%onnx::Conv_949[FLOAT, 552x1x5x5]
%onnx::Conv_952[FLOAT, 184x552x1x1]
%onnx::Conv_955[FLOAT, 1104x184x1x1]
%onnx::Conv_956[FLOAT, 1104]
%onnx::Conv_958[FLOAT, 1104x1x5x5]
%onnx::Conv_961[FLOAT, 352x1104x1x1]
%onnx::Conv_962[FLOAT, 352]
%onnx::Conv_964[FLOAT, 1504x352x1x1]
%onnx::Conv_965[FLOAT, 1504]
) {
%onnx::Conv_959 = Identity(%onnx::Conv_956)
%onnx::Conv_953 = Identity(%onnx::Conv_926)
%onnx::Conv_950 = Identity(%onnx::Conv_947)
%onnx::Conv_944 = Identity(%onnx::Conv_926)
%onnx::Conv_941 = Identity(%onnx::Conv_926)
%onnx::Conv_938 = Identity(%onnx::Conv_926)
%onnx::Conv_935 = Identity(%onnx::Conv_926)
%onnx::Conv_932 = Identity(%onnx::Conv_926)
%onnx::Conv_929 = Identity(%onnx::Conv_926)
%onnx::Conv_923 = Identity(%onnx::Conv_896)
%onnx::Conv_920 = Identity(%onnx::Conv_908)
%onnx::Conv_917 = Identity(%onnx::Conv_908)
%onnx::Conv_914 = Identity(%onnx::Conv_896)
%onnx::Conv_911 = Identity(%onnx::Conv_908)
%onnx::Conv_905 = Identity(%onnx::Conv_896)
%onnx::Conv_902 = Identity(%onnx::Conv_896)
%onnx::Conv_899 = Identity(%onnx::Conv_896)
%onnx::Conv_893 = Identity(%onnx::Conv_860)
%onnx::Conv_890 = Identity(%onnx::Conv_860)
%onnx::Conv_887 = Identity(%onnx::Conv_860)
%onnx::Conv_884 = Identity(%onnx::Conv_860)
%onnx::Conv_881 = Identity(%onnx::Conv_860)
%onnx::Conv_878 = Identity(%onnx::Conv_860)
%onnx::Conv_875 = Identity(%onnx::Conv_860)
%onnx::Conv_872 = Identity(%onnx::Conv_860)
%onnx::Conv_869 = Identity(%onnx::Conv_860)
%onnx::Conv_866 = Identity(%onnx::Conv_863)
%onnx::Conv_857 = Identity(%onnx::Conv_824)
%onnx::Conv_854 = Identity(%onnx::Conv_824)
%onnx::Conv_851 = Identity(%onnx::Conv_824)
%onnx::Conv_848 = Identity(%onnx::Conv_824)
%onnx::Conv_845 = Identity(%onnx::Conv_824)
%onnx::Conv_842 = Identity(%onnx::Conv_824)
%onnx::Conv_839 = Identity(%onnx::Conv_824)
%onnx::Conv_836 = Identity(%onnx::Conv_824)
%onnx::Conv_833 = Identity(%onnx::Conv_824)
%onnx::Conv_830 = Identity(%onnx::Conv_824)
%onnx::Conv_827 = Identity(%onnx::Conv_824)
%onnx::Conv_821 = Identity(%onnx::Conv_788)
%onnx::Conv_818 = Identity(%onnx::Conv_788)
%onnx::Conv_815 = Identity(%onnx::Conv_788)
%onnx::Conv_812 = Identity(%onnx::Conv_791)
%onnx::Conv_809 = Identity(%onnx::Conv_791)
%onnx::Conv_806 = Identity(%onnx::Conv_788)
%onnx::Conv_803 = Identity(%onnx::Conv_788)
%onnx::Conv_800 = Identity(%onnx::Conv_788)
%onnx::Conv_797 = Identity(%onnx::Conv_788)
%onnx::Conv_794 = Identity(%onnx::Conv_791)
%onnx::Conv_785 = Identity(%onnx::Conv_773)
%onnx::Conv_782 = Identity(%onnx::Conv_773)
%onnx::Conv_779 = Identity(%onnx::Conv_770)
%onnx::Conv_776 = Identity(%onnx::Conv_773)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_769, %onnx::Conv_770)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_862, %onnx::Conv_863)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_865, %onnx::Conv_866)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_868, %onnx::Conv_869)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_871, %onnx::Conv_872)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_874, %onnx::Conv_875)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_877, %onnx::Conv_878)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_880, %onnx::Conv_881)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_883, %onnx::Conv_884)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_886, %onnx::Conv_887)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_889, %onnx::Conv_890)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_892, %onnx::Conv_893)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_895, %onnx::Conv_896)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_898, %onnx::Conv_899)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_901, %onnx::Conv_902)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_904, %onnx::Conv_905)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_907, %onnx::Conv_908)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_910, %onnx::Conv_911)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_913, %onnx::Conv_914)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_916, %onnx::Conv_917)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_919, %onnx::Conv_920)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_922, %onnx::Conv_923)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_925, %onnx::Conv_926)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_928, %onnx::Conv_929)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_931, %onnx::Conv_932)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_934, %onnx::Conv_935)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_937, %onnx::Conv_938)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_940, %onnx::Conv_941)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_943, %onnx::Conv_944)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_946, %onnx::Conv_947)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_949, %onnx::Conv_950)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_952, %onnx::Conv_953)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_955, %onnx::Conv_956)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_958, %onnx::Conv_959)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_961, %onnx::Conv_962)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_964, %onnx::Conv_965)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%767 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %767
}
|
val_accuracy
| 0 | 83,540,480 | 2,095,932 |
{'zcp_synflow': 76.4964613324315, 'zcp_zen': 68.26704406738281, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.23836708068847656, 'zcp_flops': 83540480.0, 'zcp_grad_norm': 29.49294662475586, 'zcp_grasp': -0.36556243896484375, 'zcp_jacov': -16.064059658838644, 'zcp_l2_norm': 615.5392456054688, 'zcp_nwot': 215.47408816457084, 'zcp_params': 2095932.0, 'zcp_plain': -0.000730456318706274, 'zcp_snip': 46.269859313964844, 'lat_1080ti_1': 0.8238538748034633, 'lat_1080ti_32': 0.8510920959500563, 'lat_1080ti_64': 0.7736350810559876, 'lat_2080ti_1': 0.8723931789732748, 'lat_2080ti_32': 0.8887760777732101, 'lat_2080ti_64': 0.8091856560804446, 'lat_essential_ph_1': 0.39622641509433965, 'lat_eyeriss': 0.6092091027451269, 'lat_fpga': 0.6636618213123029, 'lat_gold_6226': 0.5694211928402043, 'lat_gold_6240': 0.6538567747593305, 'lat_pixel2': 0.391304347826087, 'lat_pixel3': 0.6885111600529633, 'lat_raspi4': 0.725414270491397, 'lat_samsung_a50': 0.2736842105263158, 'lat_samsung_s7': 0.2204724409448819, 'lat_silver_4114': 0.6762416198829898, 'lat_silver_4210r': 0.7459067442549236, 'lat_titan_rtx_1': 0.8174195623259443, 'lat_titan_rtx_32': 0.8361417452252116, 'lat_titan_rtx_64': 0.8324712829974089, 'lat_titanx_1': 0.43422954492776844, 'lat_titanx_32': 0.835214245125874, 'lat_titanx_64': 0.7175701105414432, 'lat_titanxp_1': 0.7574394157771749, 'lat_titanxp_32': 0.8412716806746438, 'lat_titanxp_64': 0.7929045870488274}
| |
FBNet_4530
|
FBNet
|
4530
|
4530
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_651[FLOAT, 16x3x3x3]
%onnx::Conv_652[FLOAT, 16]
%onnx::Conv_654[FLOAT, 96x16x1x1]
%onnx::Conv_655[FLOAT, 96]
%onnx::Conv_657[FLOAT, 96x1x5x5]
%onnx::Conv_660[FLOAT, 16x96x1x1]
%onnx::Conv_663[FLOAT, 96x16x1x1]
%onnx::Conv_666[FLOAT, 96x1x5x5]
%onnx::Conv_669[FLOAT, 24x96x1x1]
%onnx::Conv_670[FLOAT, 24]
%onnx::Conv_672[FLOAT, 24x12x1x1]
%onnx::Conv_675[FLOAT, 24x1x5x5]
%onnx::Conv_678[FLOAT, 24x12x1x1]
%onnx::Conv_681[FLOAT, 24x24x1x1]
%onnx::Conv_684[FLOAT, 24x1x5x5]
%onnx::Conv_687[FLOAT, 24x24x1x1]
%onnx::Conv_690[FLOAT, 24x24x1x1]
%onnx::Conv_693[FLOAT, 24x1x3x3]
%onnx::Conv_696[FLOAT, 24x24x1x1]
%onnx::Conv_699[FLOAT, 24x12x1x1]
%onnx::Conv_702[FLOAT, 24x1x5x5]
%onnx::Conv_705[FLOAT, 32x12x1x1]
%onnx::Conv_706[FLOAT, 32]
%onnx::Conv_708[FLOAT, 32x16x1x1]
%onnx::Conv_711[FLOAT, 32x1x3x3]
%onnx::Conv_714[FLOAT, 32x16x1x1]
%onnx::Conv_717[FLOAT, 32x32x1x1]
%onnx::Conv_720[FLOAT, 32x1x3x3]
%onnx::Conv_723[FLOAT, 64x32x1x1]
%onnx::Conv_724[FLOAT, 64]
%onnx::Conv_726[FLOAT, 64x32x1x1]
%onnx::Conv_729[FLOAT, 64x1x3x3]
%onnx::Conv_732[FLOAT, 64x32x1x1]
%onnx::Conv_735[FLOAT, 384x64x1x1]
%onnx::Conv_736[FLOAT, 384]
%onnx::Conv_738[FLOAT, 384x1x5x5]
%onnx::Conv_741[FLOAT, 64x384x1x1]
%onnx::Conv_744[FLOAT, 384x64x1x1]
%onnx::Conv_747[FLOAT, 384x1x5x5]
%onnx::Conv_750[FLOAT, 64x384x1x1]
%onnx::Conv_753[FLOAT, 384x64x1x1]
%onnx::Conv_756[FLOAT, 384x1x3x3]
%onnx::Conv_759[FLOAT, 112x384x1x1]
%onnx::Conv_760[FLOAT, 112]
%onnx::Conv_762[FLOAT, 672x112x1x1]
%onnx::Conv_763[FLOAT, 672]
%onnx::Conv_765[FLOAT, 672x1x5x5]
%onnx::Conv_768[FLOAT, 112x672x1x1]
%onnx::Conv_771[FLOAT, 112x56x1x1]
%onnx::Conv_774[FLOAT, 112x1x3x3]
%onnx::Conv_777[FLOAT, 112x56x1x1]
%onnx::Conv_780[FLOAT, 112x112x1x1]
%onnx::Conv_783[FLOAT, 112x1x5x5]
%onnx::Conv_786[FLOAT, 112x112x1x1]
%onnx::Conv_789[FLOAT, 336x112x1x1]
%onnx::Conv_790[FLOAT, 336]
%onnx::Conv_792[FLOAT, 336x1x5x5]
%onnx::Conv_795[FLOAT, 184x336x1x1]
%onnx::Conv_796[FLOAT, 184]
%onnx::Conv_798[FLOAT, 184x184x1x1]
%onnx::Conv_801[FLOAT, 184x1x3x3]
%onnx::Conv_804[FLOAT, 184x184x1x1]
%onnx::Conv_807[FLOAT, 184x92x1x1]
%onnx::Conv_810[FLOAT, 184x1x3x3]
%onnx::Conv_813[FLOAT, 184x92x1x1]
%onnx::Conv_816[FLOAT, 1104x184x1x1]
%onnx::Conv_817[FLOAT, 1104]
%onnx::Conv_819[FLOAT, 1104x1x5x5]
%onnx::Conv_822[FLOAT, 352x1104x1x1]
%onnx::Conv_823[FLOAT, 352]
%onnx::Conv_825[FLOAT, 1504x352x1x1]
%onnx::Conv_826[FLOAT, 1504]
) {
%onnx::Conv_820 = Identity(%onnx::Conv_817)
%onnx::Conv_814 = Identity(%onnx::Conv_796)
%onnx::Conv_811 = Identity(%onnx::Conv_796)
%onnx::Conv_808 = Identity(%onnx::Conv_796)
%onnx::Conv_805 = Identity(%onnx::Conv_796)
%onnx::Conv_802 = Identity(%onnx::Conv_796)
%onnx::Conv_799 = Identity(%onnx::Conv_796)
%onnx::Conv_793 = Identity(%onnx::Conv_790)
%onnx::Conv_787 = Identity(%onnx::Conv_760)
%onnx::Conv_784 = Identity(%onnx::Conv_760)
%onnx::Conv_781 = Identity(%onnx::Conv_760)
%onnx::Conv_778 = Identity(%onnx::Conv_760)
%onnx::Conv_775 = Identity(%onnx::Conv_760)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_763)
%onnx::Conv_757 = Identity(%onnx::Conv_736)
%onnx::Conv_754 = Identity(%onnx::Conv_736)
%onnx::Conv_751 = Identity(%onnx::Conv_724)
%onnx::Conv_748 = Identity(%onnx::Conv_736)
%onnx::Conv_745 = Identity(%onnx::Conv_736)
%onnx::Conv_742 = Identity(%onnx::Conv_724)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_733 = Identity(%onnx::Conv_724)
%onnx::Conv_730 = Identity(%onnx::Conv_724)
%onnx::Conv_727 = Identity(%onnx::Conv_724)
%onnx::Conv_721 = Identity(%onnx::Conv_706)
%onnx::Conv_718 = Identity(%onnx::Conv_706)
%onnx::Conv_715 = Identity(%onnx::Conv_706)
%onnx::Conv_712 = Identity(%onnx::Conv_706)
%onnx::Conv_709 = Identity(%onnx::Conv_706)
%onnx::Conv_703 = Identity(%onnx::Conv_670)
%onnx::Conv_700 = Identity(%onnx::Conv_670)
%onnx::Conv_697 = Identity(%onnx::Conv_670)
%onnx::Conv_694 = Identity(%onnx::Conv_670)
%onnx::Conv_691 = Identity(%onnx::Conv_670)
%onnx::Conv_688 = Identity(%onnx::Conv_670)
%onnx::Conv_685 = Identity(%onnx::Conv_670)
%onnx::Conv_682 = Identity(%onnx::Conv_670)
%onnx::Conv_679 = Identity(%onnx::Conv_670)
%onnx::Conv_676 = Identity(%onnx::Conv_670)
%onnx::Conv_673 = Identity(%onnx::Conv_670)
%onnx::Conv_667 = Identity(%onnx::Conv_655)
%onnx::Conv_664 = Identity(%onnx::Conv_655)
%onnx::Conv_661 = Identity(%onnx::Conv_652)
%onnx::Conv_658 = Identity(%onnx::Conv_655)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_651, %onnx::Conv_652)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%649 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %649
}
|
val_accuracy
| 0 | 70,458,496 | 1,960,052 |
{'zcp_synflow': 71.6572627554298, 'zcp_zen': 63.552879333496094, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.13901016116142273, 'zcp_flops': 70458496.0, 'zcp_grad_norm': 25.314590454101562, 'zcp_grasp': -0.22984695434570312, 'zcp_jacov': -16.05331191697255, 'zcp_l2_norm': 578.187744140625, 'zcp_nwot': 211.1612872173845, 'zcp_params': 1960052.0, 'zcp_plain': 0.002896197373047471, 'zcp_snip': 39.78757095336914, 'lat_1080ti_1': 0.5032149905276254, 'lat_1080ti_32': 0.5202286293462799, 'lat_1080ti_64': 0.45134944554917, 'lat_2080ti_1': 0.5628270070136282, 'lat_2080ti_32': 0.5546351712658925, 'lat_2080ti_64': 0.4476952975899689, 'lat_essential_ph_1': 0.32075471698113206, 'lat_eyeriss': 0.4672950323657237, 'lat_fpga': 0.4813920581046715, 'lat_gold_6226': 0.3802551649180215, 'lat_gold_6240': 0.5679288446695591, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.5312103769857802, 'lat_raspi4': 0.551550812996304, 'lat_samsung_a50': 0.18947368421052632, 'lat_samsung_s7': 0.1732283464566929, 'lat_silver_4114': 0.6670984358271848, 'lat_silver_4210r': 0.6592730910768106, 'lat_titan_rtx_1': 0.5213208046422232, 'lat_titan_rtx_32': 0.5291345642755116, 'lat_titan_rtx_64': 0.4568792979518929, 'lat_titanx_1': 0.2762127944806117, 'lat_titanx_32': 0.4729353382699656, 'lat_titanx_64': 0.4582305403866523, 'lat_titanxp_1': 0.47998431281440257, 'lat_titanxp_32': 0.49768955917500424, 'lat_titanxp_64': 0.4632037686317474}
| |
FBNet_2075
|
FBNet
|
2075
|
2075
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_622[FLOAT, 16x3x3x3]
%onnx::Conv_623[FLOAT, 16]
%onnx::Conv_625[FLOAT, 16x16x1x1]
%onnx::Conv_628[FLOAT, 16x1x5x5]
%onnx::Conv_631[FLOAT, 16x16x1x1]
%onnx::Conv_634[FLOAT, 24x16x1x1]
%onnx::Conv_635[FLOAT, 24]
%onnx::Conv_637[FLOAT, 144x24x1x1]
%onnx::Conv_638[FLOAT, 144]
%onnx::Conv_640[FLOAT, 144x1x3x3]
%onnx::Conv_643[FLOAT, 24x144x1x1]
%onnx::Conv_646[FLOAT, 24x24x1x1]
%onnx::Conv_649[FLOAT, 24x1x5x5]
%onnx::Conv_652[FLOAT, 24x24x1x1]
%onnx::Conv_655[FLOAT, 24x24x1x1]
%onnx::Conv_658[FLOAT, 24x1x5x5]
%onnx::Conv_661[FLOAT, 32x24x1x1]
%onnx::Conv_662[FLOAT, 32]
%onnx::Conv_664[FLOAT, 32x32x1x1]
%onnx::Conv_667[FLOAT, 32x1x3x3]
%onnx::Conv_670[FLOAT, 32x32x1x1]
%onnx::Conv_673[FLOAT, 32x16x1x1]
%onnx::Conv_676[FLOAT, 32x1x5x5]
%onnx::Conv_679[FLOAT, 32x16x1x1]
%onnx::Conv_682[FLOAT, 32x32x1x1]
%onnx::Conv_685[FLOAT, 32x1x3x3]
%onnx::Conv_688[FLOAT, 32x32x1x1]
%onnx::Conv_691[FLOAT, 96x32x1x1]
%onnx::Conv_692[FLOAT, 96]
%onnx::Conv_694[FLOAT, 96x1x5x5]
%onnx::Conv_697[FLOAT, 64x96x1x1]
%onnx::Conv_698[FLOAT, 64]
%onnx::Conv_700[FLOAT, 384x64x1x1]
%onnx::Conv_701[FLOAT, 384]
%onnx::Conv_703[FLOAT, 384x1x5x5]
%onnx::Conv_706[FLOAT, 64x384x1x1]
%onnx::Conv_709[FLOAT, 64x64x1x1]
%onnx::Conv_712[FLOAT, 64x1x5x5]
%onnx::Conv_715[FLOAT, 64x64x1x1]
%onnx::Conv_718[FLOAT, 384x64x1x1]
%onnx::Conv_721[FLOAT, 384x1x5x5]
%onnx::Conv_724[FLOAT, 112x384x1x1]
%onnx::Conv_725[FLOAT, 112]
%onnx::Conv_727[FLOAT, 112x56x1x1]
%onnx::Conv_730[FLOAT, 112x1x3x3]
%onnx::Conv_733[FLOAT, 112x56x1x1]
%onnx::Conv_736[FLOAT, 112x56x1x1]
%onnx::Conv_739[FLOAT, 112x1x3x3]
%onnx::Conv_742[FLOAT, 112x56x1x1]
%onnx::Conv_745[FLOAT, 112x56x1x1]
%onnx::Conv_748[FLOAT, 112x1x3x3]
%onnx::Conv_751[FLOAT, 112x56x1x1]
%onnx::Conv_754[FLOAT, 112x112x1x1]
%onnx::Conv_757[FLOAT, 112x1x3x3]
%onnx::Conv_760[FLOAT, 184x112x1x1]
%onnx::Conv_761[FLOAT, 184]
%onnx::Conv_763[FLOAT, 1104x184x1x1]
%onnx::Conv_764[FLOAT, 1104]
%onnx::Conv_766[FLOAT, 1104x1x5x5]
%onnx::Conv_769[FLOAT, 184x1104x1x1]
%onnx::Conv_772[FLOAT, 1104x184x1x1]
%onnx::Conv_775[FLOAT, 1104x1x3x3]
%onnx::Conv_778[FLOAT, 184x1104x1x1]
%onnx::Conv_781[FLOAT, 184x184x1x1]
%onnx::Conv_784[FLOAT, 184x1x5x5]
%onnx::Conv_787[FLOAT, 184x184x1x1]
%onnx::Conv_790[FLOAT, 184x184x1x1]
%onnx::Conv_793[FLOAT, 184x1x5x5]
%onnx::Conv_796[FLOAT, 352x184x1x1]
%onnx::Conv_797[FLOAT, 352]
%onnx::Conv_799[FLOAT, 1504x352x1x1]
%onnx::Conv_800[FLOAT, 1504]
) {
%onnx::Conv_794 = Identity(%onnx::Conv_761)
%onnx::Conv_791 = Identity(%onnx::Conv_761)
%onnx::Conv_788 = Identity(%onnx::Conv_761)
%onnx::Conv_785 = Identity(%onnx::Conv_761)
%onnx::Conv_782 = Identity(%onnx::Conv_761)
%onnx::Conv_779 = Identity(%onnx::Conv_761)
%onnx::Conv_776 = Identity(%onnx::Conv_764)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_761)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_758 = Identity(%onnx::Conv_725)
%onnx::Conv_755 = Identity(%onnx::Conv_725)
%onnx::Conv_752 = Identity(%onnx::Conv_725)
%onnx::Conv_749 = Identity(%onnx::Conv_725)
%onnx::Conv_746 = Identity(%onnx::Conv_725)
%onnx::Conv_743 = Identity(%onnx::Conv_725)
%onnx::Conv_740 = Identity(%onnx::Conv_725)
%onnx::Conv_737 = Identity(%onnx::Conv_725)
%onnx::Conv_734 = Identity(%onnx::Conv_725)
%onnx::Conv_731 = Identity(%onnx::Conv_725)
%onnx::Conv_728 = Identity(%onnx::Conv_725)
%onnx::Conv_722 = Identity(%onnx::Conv_701)
%onnx::Conv_719 = Identity(%onnx::Conv_701)
%onnx::Conv_716 = Identity(%onnx::Conv_698)
%onnx::Conv_713 = Identity(%onnx::Conv_698)
%onnx::Conv_710 = Identity(%onnx::Conv_698)
%onnx::Conv_707 = Identity(%onnx::Conv_698)
%onnx::Conv_704 = Identity(%onnx::Conv_701)
%onnx::Conv_695 = Identity(%onnx::Conv_692)
%onnx::Conv_689 = Identity(%onnx::Conv_662)
%onnx::Conv_686 = Identity(%onnx::Conv_662)
%onnx::Conv_683 = Identity(%onnx::Conv_662)
%onnx::Conv_680 = Identity(%onnx::Conv_662)
%onnx::Conv_677 = Identity(%onnx::Conv_662)
%onnx::Conv_674 = Identity(%onnx::Conv_662)
%onnx::Conv_671 = Identity(%onnx::Conv_662)
%onnx::Conv_668 = Identity(%onnx::Conv_662)
%onnx::Conv_665 = Identity(%onnx::Conv_662)
%onnx::Conv_659 = Identity(%onnx::Conv_635)
%onnx::Conv_656 = Identity(%onnx::Conv_635)
%onnx::Conv_653 = Identity(%onnx::Conv_635)
%onnx::Conv_650 = Identity(%onnx::Conv_635)
%onnx::Conv_647 = Identity(%onnx::Conv_635)
%onnx::Conv_644 = Identity(%onnx::Conv_635)
%onnx::Conv_641 = Identity(%onnx::Conv_638)
%onnx::Conv_632 = Identity(%onnx::Conv_623)
%onnx::Conv_629 = Identity(%onnx::Conv_623)
%onnx::Conv_626 = Identity(%onnx::Conv_623)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_622, %onnx::Conv_623)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.1/relu/Relu_output_0 = Relu(%/cells.1/conv/Conv_output_0)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/relu/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/relu/Relu_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%620 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %620
}
|
val_accuracy
| 0 | 56,344,960 | 1,981,764 |
{'zcp_synflow': 75.79197504674478, 'zcp_zen': 65.18639373779297, 'zcp_epe_nas': 5.838981649282895, 'zcp_fisher': 0.09937311708927155, 'zcp_flops': 56344960.0, 'zcp_grad_norm': 23.05716323852539, 'zcp_grasp': -0.06836128234863281, 'zcp_jacov': -16.05851043007763, 'zcp_l2_norm': 592.0376586914062, 'zcp_nwot': 207.60441392305881, 'zcp_params': 1981764.0, 'zcp_plain': 0.0026447372511029243, 'zcp_snip': 37.30038833618164, 'lat_1080ti_1': 0.521945007499597, 'lat_1080ti_32': 0.47153311240410006, 'lat_1080ti_64': 0.3293491329434381, 'lat_2080ti_1': 0.5267086527769381, 'lat_2080ti_32': 0.4785876821461105, 'lat_2080ti_64': 0.32186631943155364, 'lat_essential_ph_1': 0.2641509433962264, 'lat_eyeriss': 0.32426237324189017, 'lat_fpga': 0.3337060528997911, 'lat_gold_6226': 0.2875976468539273, 'lat_gold_6240': 0.5158421444873476, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.2795858466113591, 'lat_raspi4': 0.35105795308057147, 'lat_samsung_a50': 0.14736842105263157, 'lat_samsung_s7': 0.14960629921259844, 'lat_silver_4114': 0.4990274078930101, 'lat_silver_4210r': 0.49000487773438484, 'lat_titan_rtx_1': 0.505677377882639, 'lat_titan_rtx_32': 0.46533130486934626, 'lat_titan_rtx_64': 0.3544961736033191, 'lat_titanx_1': 0.26389894317031326, 'lat_titanx_32': 0.4062272360052304, 'lat_titanx_64': 0.29837417502593044, 'lat_titanxp_1': 0.4875906811443836, 'lat_titanxp_32': 0.4393180962674308, 'lat_titanxp_64': 0.32921680051199564}
| |
FBNet_307
|
FBNet
|
307
|
307
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_613[FLOAT, 16x3x3x3]
%onnx::Conv_614[FLOAT, 16]
%onnx::Conv_616[FLOAT, 96x16x1x1]
%onnx::Conv_617[FLOAT, 96]
%onnx::Conv_619[FLOAT, 96x1x3x3]
%onnx::Conv_622[FLOAT, 16x96x1x1]
%onnx::Conv_625[FLOAT, 24x16x1x1]
%onnx::Conv_626[FLOAT, 24]
%onnx::Conv_628[FLOAT, 24x12x1x1]
%onnx::Conv_631[FLOAT, 24x1x5x5]
%onnx::Conv_634[FLOAT, 24x12x1x1]
%onnx::Conv_637[FLOAT, 24x24x1x1]
%onnx::Conv_640[FLOAT, 24x1x5x5]
%onnx::Conv_643[FLOAT, 24x24x1x1]
%onnx::Conv_646[FLOAT, 24x24x1x1]
%onnx::Conv_649[FLOAT, 24x1x3x3]
%onnx::Conv_652[FLOAT, 32x24x1x1]
%onnx::Conv_653[FLOAT, 32]
%onnx::Conv_655[FLOAT, 96x32x1x1]
%onnx::Conv_658[FLOAT, 96x1x3x3]
%onnx::Conv_661[FLOAT, 32x96x1x1]
%onnx::Conv_664[FLOAT, 192x32x1x1]
%onnx::Conv_665[FLOAT, 192]
%onnx::Conv_667[FLOAT, 192x1x3x3]
%onnx::Conv_670[FLOAT, 32x192x1x1]
%onnx::Conv_673[FLOAT, 192x32x1x1]
%onnx::Conv_676[FLOAT, 192x1x5x5]
%onnx::Conv_679[FLOAT, 32x192x1x1]
%onnx::Conv_682[FLOAT, 32x32x1x1]
%onnx::Conv_685[FLOAT, 32x1x3x3]
%onnx::Conv_688[FLOAT, 64x32x1x1]
%onnx::Conv_689[FLOAT, 64]
%onnx::Conv_691[FLOAT, 64x64x1x1]
%onnx::Conv_694[FLOAT, 64x1x5x5]
%onnx::Conv_697[FLOAT, 64x64x1x1]
%onnx::Conv_700[FLOAT, 64x64x1x1]
%onnx::Conv_703[FLOAT, 64x1x5x5]
%onnx::Conv_706[FLOAT, 64x64x1x1]
%onnx::Conv_709[FLOAT, 384x64x1x1]
%onnx::Conv_710[FLOAT, 384]
%onnx::Conv_712[FLOAT, 384x1x3x3]
%onnx::Conv_715[FLOAT, 64x384x1x1]
%onnx::Conv_718[FLOAT, 64x32x1x1]
%onnx::Conv_721[FLOAT, 64x1x3x3]
%onnx::Conv_724[FLOAT, 112x32x1x1]
%onnx::Conv_725[FLOAT, 112]
%onnx::Conv_727[FLOAT, 112x112x1x1]
%onnx::Conv_730[FLOAT, 112x1x3x3]
%onnx::Conv_733[FLOAT, 112x112x1x1]
%onnx::Conv_736[FLOAT, 672x112x1x1]
%onnx::Conv_737[FLOAT, 672]
%onnx::Conv_739[FLOAT, 672x1x5x5]
%onnx::Conv_742[FLOAT, 112x672x1x1]
%onnx::Conv_745[FLOAT, 672x112x1x1]
%onnx::Conv_748[FLOAT, 672x1x5x5]
%onnx::Conv_751[FLOAT, 112x672x1x1]
%onnx::Conv_754[FLOAT, 112x112x1x1]
%onnx::Conv_757[FLOAT, 112x1x5x5]
%onnx::Conv_760[FLOAT, 184x112x1x1]
%onnx::Conv_761[FLOAT, 184]
%onnx::Conv_763[FLOAT, 1104x184x1x1]
%onnx::Conv_764[FLOAT, 1104]
%onnx::Conv_766[FLOAT, 1104x1x5x5]
%onnx::Conv_769[FLOAT, 184x1104x1x1]
%onnx::Conv_772[FLOAT, 1104x184x1x1]
%onnx::Conv_775[FLOAT, 1104x1x3x3]
%onnx::Conv_778[FLOAT, 184x1104x1x1]
%onnx::Conv_781[FLOAT, 552x184x1x1]
%onnx::Conv_782[FLOAT, 552]
%onnx::Conv_784[FLOAT, 552x1x5x5]
%onnx::Conv_787[FLOAT, 184x552x1x1]
%onnx::Conv_790[FLOAT, 184x184x1x1]
%onnx::Conv_793[FLOAT, 184x1x5x5]
%onnx::Conv_796[FLOAT, 352x184x1x1]
%onnx::Conv_797[FLOAT, 352]
%onnx::Conv_799[FLOAT, 1504x352x1x1]
%onnx::Conv_800[FLOAT, 1504]
) {
%onnx::Conv_794 = Identity(%onnx::Conv_761)
%onnx::Conv_791 = Identity(%onnx::Conv_761)
%onnx::Conv_788 = Identity(%onnx::Conv_761)
%onnx::Conv_785 = Identity(%onnx::Conv_782)
%onnx::Conv_779 = Identity(%onnx::Conv_761)
%onnx::Conv_776 = Identity(%onnx::Conv_764)
%onnx::Conv_773 = Identity(%onnx::Conv_764)
%onnx::Conv_770 = Identity(%onnx::Conv_761)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_758 = Identity(%onnx::Conv_725)
%onnx::Conv_755 = Identity(%onnx::Conv_725)
%onnx::Conv_752 = Identity(%onnx::Conv_725)
%onnx::Conv_749 = Identity(%onnx::Conv_737)
%onnx::Conv_746 = Identity(%onnx::Conv_737)
%onnx::Conv_743 = Identity(%onnx::Conv_725)
%onnx::Conv_740 = Identity(%onnx::Conv_737)
%onnx::Conv_734 = Identity(%onnx::Conv_725)
%onnx::Conv_731 = Identity(%onnx::Conv_725)
%onnx::Conv_728 = Identity(%onnx::Conv_725)
%onnx::Conv_722 = Identity(%onnx::Conv_689)
%onnx::Conv_719 = Identity(%onnx::Conv_689)
%onnx::Conv_716 = Identity(%onnx::Conv_689)
%onnx::Conv_713 = Identity(%onnx::Conv_710)
%onnx::Conv_707 = Identity(%onnx::Conv_689)
%onnx::Conv_704 = Identity(%onnx::Conv_689)
%onnx::Conv_701 = Identity(%onnx::Conv_689)
%onnx::Conv_698 = Identity(%onnx::Conv_689)
%onnx::Conv_695 = Identity(%onnx::Conv_689)
%onnx::Conv_692 = Identity(%onnx::Conv_689)
%onnx::Conv_686 = Identity(%onnx::Conv_653)
%onnx::Conv_683 = Identity(%onnx::Conv_653)
%onnx::Conv_680 = Identity(%onnx::Conv_653)
%onnx::Conv_677 = Identity(%onnx::Conv_665)
%onnx::Conv_674 = Identity(%onnx::Conv_665)
%onnx::Conv_671 = Identity(%onnx::Conv_653)
%onnx::Conv_668 = Identity(%onnx::Conv_665)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_617)
%onnx::Conv_656 = Identity(%onnx::Conv_617)
%onnx::Conv_650 = Identity(%onnx::Conv_626)
%onnx::Conv_647 = Identity(%onnx::Conv_626)
%onnx::Conv_644 = Identity(%onnx::Conv_626)
%onnx::Conv_641 = Identity(%onnx::Conv_626)
%onnx::Conv_638 = Identity(%onnx::Conv_626)
%onnx::Conv_635 = Identity(%onnx::Conv_626)
%onnx::Conv_632 = Identity(%onnx::Conv_626)
%onnx::Conv_629 = Identity(%onnx::Conv_626)
%onnx::Conv_623 = Identity(%onnx::Conv_614)
%onnx::Conv_620 = Identity(%onnx::Conv_617)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_613, %onnx::Conv_614)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.1/relu/Relu_output_0 = Relu(%/cells.1/conv/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/relu/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/relu/Relu_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%611 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %611
}
|
val_accuracy
| 0 | 78,390,144 | 2,406,092 |
{'zcp_synflow': 81.99246909300852, 'zcp_zen': 72.36370849609375, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.1224769800901413, 'zcp_flops': 78390144.0, 'zcp_grad_norm': 26.32333755493164, 'zcp_grasp': -0.33087158203125, 'zcp_jacov': -16.057862992234547, 'zcp_l2_norm': 699.5054321289062, 'zcp_nwot': 211.54754880656606, 'zcp_params': 2406092.0, 'zcp_plain': -0.00032716355053707957, 'zcp_snip': 43.68217468261719, 'lat_1080ti_1': 0.5496001598084274, 'lat_1080ti_32': 0.49274889840087127, 'lat_1080ti_64': 0.39264857551686083, 'lat_2080ti_1': 0.5614201037545018, 'lat_2080ti_32': 0.46819938274230916, 'lat_2080ti_64': 0.3941013174937289, 'lat_essential_ph_1': 0.49056603773584906, 'lat_eyeriss': 0.5665560302935839, 'lat_fpga': 0.635549530472298, 'lat_gold_6226': 0.5129135000623991, 'lat_gold_6240': 0.7172629453463325, 'lat_pixel2': 0.5217391304347826, 'lat_pixel3': 0.5235795160782983, 'lat_raspi4': 0.5301745166004816, 'lat_samsung_a50': 0.25263157894736843, 'lat_samsung_s7': 0.29133858267716534, 'lat_silver_4114': 0.8647886447307325, 'lat_silver_4210r': 0.7458453383461864, 'lat_titan_rtx_1': 0.554338534784858, 'lat_titan_rtx_32': 0.4760997123320076, 'lat_titan_rtx_64': 0.4056907753887331, 'lat_titanx_1': 0.3088985633459277, 'lat_titanx_32': 0.40435847597900404, 'lat_titanx_64': 0.38418345634006124, 'lat_titanxp_1': 0.5495714566107898, 'lat_titanxp_32': 0.4672000186019189, 'lat_titanxp_64': 0.3978509786954406}
| |
FBNet_73
|
FBNet
|
73
|
73
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_569[FLOAT, 16x3x3x3]
%onnx::Conv_570[FLOAT, 16]
%onnx::Conv_572[FLOAT, 16x8x1x1]
%onnx::Conv_575[FLOAT, 16x1x3x3]
%onnx::Conv_578[FLOAT, 16x8x1x1]
%onnx::Conv_581[FLOAT, 48x16x1x1]
%onnx::Conv_582[FLOAT, 48]
%onnx::Conv_584[FLOAT, 48x1x5x5]
%onnx::Conv_587[FLOAT, 24x48x1x1]
%onnx::Conv_588[FLOAT, 24]
%onnx::Conv_590[FLOAT, 24x24x1x1]
%onnx::Conv_593[FLOAT, 24x1x3x3]
%onnx::Conv_596[FLOAT, 24x24x1x1]
%onnx::Conv_599[FLOAT, 24x24x1x1]
%onnx::Conv_602[FLOAT, 24x1x5x5]
%onnx::Conv_605[FLOAT, 24x24x1x1]
%onnx::Conv_608[FLOAT, 144x24x1x1]
%onnx::Conv_609[FLOAT, 144]
%onnx::Conv_611[FLOAT, 144x1x5x5]
%onnx::Conv_614[FLOAT, 32x144x1x1]
%onnx::Conv_615[FLOAT, 32]
%onnx::Conv_617[FLOAT, 32x16x1x1]
%onnx::Conv_620[FLOAT, 32x1x3x3]
%onnx::Conv_623[FLOAT, 32x16x1x1]
%onnx::Conv_626[FLOAT, 32x32x1x1]
%onnx::Conv_629[FLOAT, 32x1x5x5]
%onnx::Conv_632[FLOAT, 32x32x1x1]
%onnx::Conv_635[FLOAT, 192x32x1x1]
%onnx::Conv_636[FLOAT, 192]
%onnx::Conv_638[FLOAT, 192x1x5x5]
%onnx::Conv_641[FLOAT, 64x192x1x1]
%onnx::Conv_642[FLOAT, 64]
%onnx::Conv_644[FLOAT, 384x64x1x1]
%onnx::Conv_645[FLOAT, 384]
%onnx::Conv_647[FLOAT, 384x1x3x3]
%onnx::Conv_650[FLOAT, 64x384x1x1]
%onnx::Conv_653[FLOAT, 64x32x1x1]
%onnx::Conv_656[FLOAT, 64x1x5x5]
%onnx::Conv_659[FLOAT, 64x32x1x1]
%onnx::Conv_662[FLOAT, 64x32x1x1]
%onnx::Conv_665[FLOAT, 64x1x5x5]
%onnx::Conv_668[FLOAT, 64x32x1x1]
%onnx::Conv_671[FLOAT, 64x64x1x1]
%onnx::Conv_674[FLOAT, 64x1x5x5]
%onnx::Conv_677[FLOAT, 112x64x1x1]
%onnx::Conv_678[FLOAT, 112]
%onnx::Conv_680[FLOAT, 672x112x1x1]
%onnx::Conv_681[FLOAT, 672]
%onnx::Conv_683[FLOAT, 672x1x5x5]
%onnx::Conv_686[FLOAT, 112x672x1x1]
%onnx::Conv_689[FLOAT, 336x112x1x1]
%onnx::Conv_690[FLOAT, 336]
%onnx::Conv_692[FLOAT, 336x1x3x3]
%onnx::Conv_695[FLOAT, 112x336x1x1]
%onnx::Conv_698[FLOAT, 336x112x1x1]
%onnx::Conv_701[FLOAT, 336x1x3x3]
%onnx::Conv_704[FLOAT, 112x336x1x1]
%onnx::Conv_707[FLOAT, 184x112x1x1]
%onnx::Conv_708[FLOAT, 184]
%onnx::Conv_710[FLOAT, 1104x184x1x1]
%onnx::Conv_711[FLOAT, 1104]
%onnx::Conv_713[FLOAT, 1104x1x3x3]
%onnx::Conv_716[FLOAT, 184x1104x1x1]
%onnx::Conv_719[FLOAT, 552x184x1x1]
%onnx::Conv_720[FLOAT, 552]
%onnx::Conv_722[FLOAT, 552x1x5x5]
%onnx::Conv_725[FLOAT, 352x552x1x1]
%onnx::Conv_726[FLOAT, 352]
%onnx::Conv_728[FLOAT, 1504x352x1x1]
%onnx::Conv_729[FLOAT, 1504]
) {
%onnx::Conv_723 = Identity(%onnx::Conv_720)
%onnx::Conv_717 = Identity(%onnx::Conv_708)
%onnx::Conv_714 = Identity(%onnx::Conv_711)
%onnx::Conv_705 = Identity(%onnx::Conv_678)
%onnx::Conv_702 = Identity(%onnx::Conv_690)
%onnx::Conv_699 = Identity(%onnx::Conv_690)
%onnx::Conv_696 = Identity(%onnx::Conv_678)
%onnx::Conv_693 = Identity(%onnx::Conv_690)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_675 = Identity(%onnx::Conv_642)
%onnx::Conv_672 = Identity(%onnx::Conv_642)
%onnx::Conv_669 = Identity(%onnx::Conv_642)
%onnx::Conv_666 = Identity(%onnx::Conv_642)
%onnx::Conv_663 = Identity(%onnx::Conv_642)
%onnx::Conv_660 = Identity(%onnx::Conv_642)
%onnx::Conv_657 = Identity(%onnx::Conv_642)
%onnx::Conv_654 = Identity(%onnx::Conv_642)
%onnx::Conv_651 = Identity(%onnx::Conv_642)
%onnx::Conv_648 = Identity(%onnx::Conv_645)
%onnx::Conv_639 = Identity(%onnx::Conv_636)
%onnx::Conv_633 = Identity(%onnx::Conv_615)
%onnx::Conv_630 = Identity(%onnx::Conv_615)
%onnx::Conv_627 = Identity(%onnx::Conv_615)
%onnx::Conv_624 = Identity(%onnx::Conv_615)
%onnx::Conv_621 = Identity(%onnx::Conv_615)
%onnx::Conv_618 = Identity(%onnx::Conv_615)
%onnx::Conv_612 = Identity(%onnx::Conv_609)
%onnx::Conv_606 = Identity(%onnx::Conv_588)
%onnx::Conv_603 = Identity(%onnx::Conv_588)
%onnx::Conv_600 = Identity(%onnx::Conv_588)
%onnx::Conv_597 = Identity(%onnx::Conv_588)
%onnx::Conv_594 = Identity(%onnx::Conv_588)
%onnx::Conv_591 = Identity(%onnx::Conv_588)
%onnx::Conv_585 = Identity(%onnx::Conv_582)
%onnx::Conv_579 = Identity(%onnx::Conv_570)
%onnx::Conv_576 = Identity(%onnx::Conv_570)
%onnx::Conv_573 = Identity(%onnx::Conv_570)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_569, %onnx::Conv_570)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_572, %onnx::Conv_573)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_575, %onnx::Conv_576)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_578, %onnx::Conv_579)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_581, %onnx::Conv_582)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_584, %onnx::Conv_585)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_587, %onnx::Conv_588)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_590, %onnx::Conv_591)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_593, %onnx::Conv_594)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_596, %onnx::Conv_597)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_599, %onnx::Conv_600)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_602, %onnx::Conv_603)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_605, %onnx::Conv_606)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_608, %onnx::Conv_609)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_611, %onnx::Conv_612)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_614, %onnx::Conv_615)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_617, %onnx::Conv_618)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_620, %onnx::Conv_621)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_623, %onnx::Conv_624)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_626, %onnx::Conv_627)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_629, %onnx::Conv_630)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_632, %onnx::Conv_633)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_635, %onnx::Conv_636)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_638, %onnx::Conv_639)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_641, %onnx::Conv_642)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%567 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %567
}
|
val_accuracy
| 0 | 65,261,056 | 1,895,900 |
{'zcp_synflow': 68.93118905320381, 'zcp_zen': 59.017578125, 'zcp_epe_nas': 15.877983958431974, 'zcp_fisher': 0.04457683861255646, 'zcp_flops': 65261056.0, 'zcp_grad_norm': 18.76033592224121, 'zcp_grasp': -0.015143394470214844, 'zcp_jacov': -16.05013842333023, 'zcp_l2_norm': 544.3856811523438, 'zcp_nwot': 209.00296141363907, 'zcp_params': 1895900.0, 'zcp_plain': 0.0030477249529212713, 'zcp_snip': 30.62595558166504, 'lat_1080ti_1': 0.26611112971051293, 'lat_1080ti_32': 0.21176407437177902, 'lat_1080ti_64': 0.22917624165090558, 'lat_2080ti_1': 0.30212216237557993, 'lat_2080ti_32': 0.25393381686615607, 'lat_2080ti_64': 0.21849663932035726, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.328324134010599, 'lat_fpga': 0.41085900315657176, 'lat_gold_6226': 0.3588636433175228, 'lat_gold_6240': 0.4656135627319631, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.3550199405437249, 'lat_raspi4': 0.4434324355214971, 'lat_samsung_a50': 0.14736842105263157, 'lat_samsung_s7': 0.12598425196850394, 'lat_silver_4114': 0.7130825187066979, 'lat_silver_4210r': 0.3618548213709782, 'lat_titan_rtx_1': 0.3016433498290347, 'lat_titan_rtx_32': 0.254287077193095, 'lat_titan_rtx_64': 0.21094527821717213, 'lat_titanx_1': 0.16640008615294416, 'lat_titanx_32': 0.21662112659904753, 'lat_titanx_64': 0.21397534738998186, 'lat_titanxp_1': 0.3083583111254409, 'lat_titanxp_32': 0.2339317358887199, 'lat_titanxp_64': 0.2269141214016377}
| |
FBNet_1523
|
FBNet
|
1523
|
1523
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_667[FLOAT, 16x3x3x3]
%onnx::Conv_668[FLOAT, 16]
%onnx::Conv_670[FLOAT, 16x8x1x1]
%onnx::Conv_673[FLOAT, 16x1x3x3]
%onnx::Conv_676[FLOAT, 16x8x1x1]
%onnx::Conv_679[FLOAT, 96x16x1x1]
%onnx::Conv_680[FLOAT, 96]
%onnx::Conv_682[FLOAT, 96x1x5x5]
%onnx::Conv_685[FLOAT, 24x96x1x1]
%onnx::Conv_686[FLOAT, 24]
%onnx::Conv_688[FLOAT, 24x12x1x1]
%onnx::Conv_691[FLOAT, 24x1x5x5]
%onnx::Conv_694[FLOAT, 24x12x1x1]
%onnx::Conv_697[FLOAT, 72x24x1x1]
%onnx::Conv_698[FLOAT, 72]
%onnx::Conv_700[FLOAT, 72x1x5x5]
%onnx::Conv_703[FLOAT, 24x72x1x1]
%onnx::Conv_706[FLOAT, 144x24x1x1]
%onnx::Conv_707[FLOAT, 144]
%onnx::Conv_709[FLOAT, 144x1x5x5]
%onnx::Conv_712[FLOAT, 24x144x1x1]
%onnx::Conv_715[FLOAT, 24x24x1x1]
%onnx::Conv_718[FLOAT, 24x1x3x3]
%onnx::Conv_721[FLOAT, 32x24x1x1]
%onnx::Conv_722[FLOAT, 32]
%onnx::Conv_724[FLOAT, 32x32x1x1]
%onnx::Conv_727[FLOAT, 32x1x5x5]
%onnx::Conv_730[FLOAT, 32x32x1x1]
%onnx::Conv_733[FLOAT, 192x32x1x1]
%onnx::Conv_734[FLOAT, 192]
%onnx::Conv_736[FLOAT, 192x1x5x5]
%onnx::Conv_739[FLOAT, 32x192x1x1]
%onnx::Conv_742[FLOAT, 32x32x1x1]
%onnx::Conv_745[FLOAT, 32x1x3x3]
%onnx::Conv_748[FLOAT, 32x32x1x1]
%onnx::Conv_751[FLOAT, 32x32x1x1]
%onnx::Conv_754[FLOAT, 32x1x3x3]
%onnx::Conv_757[FLOAT, 64x32x1x1]
%onnx::Conv_758[FLOAT, 64]
%onnx::Conv_760[FLOAT, 64x32x1x1]
%onnx::Conv_763[FLOAT, 64x1x3x3]
%onnx::Conv_766[FLOAT, 64x32x1x1]
%onnx::Conv_769[FLOAT, 384x64x1x1]
%onnx::Conv_770[FLOAT, 384]
%onnx::Conv_772[FLOAT, 384x1x5x5]
%onnx::Conv_775[FLOAT, 64x384x1x1]
%onnx::Conv_778[FLOAT, 64x32x1x1]
%onnx::Conv_781[FLOAT, 64x1x5x5]
%onnx::Conv_784[FLOAT, 64x32x1x1]
%onnx::Conv_787[FLOAT, 384x64x1x1]
%onnx::Conv_790[FLOAT, 384x1x5x5]
%onnx::Conv_793[FLOAT, 112x384x1x1]
%onnx::Conv_794[FLOAT, 112]
%onnx::Conv_796[FLOAT, 336x112x1x1]
%onnx::Conv_797[FLOAT, 336]
%onnx::Conv_799[FLOAT, 336x1x5x5]
%onnx::Conv_802[FLOAT, 112x336x1x1]
%onnx::Conv_805[FLOAT, 672x112x1x1]
%onnx::Conv_806[FLOAT, 672]
%onnx::Conv_808[FLOAT, 672x1x5x5]
%onnx::Conv_811[FLOAT, 112x672x1x1]
%onnx::Conv_814[FLOAT, 112x112x1x1]
%onnx::Conv_817[FLOAT, 112x1x3x3]
%onnx::Conv_820[FLOAT, 184x112x1x1]
%onnx::Conv_821[FLOAT, 184]
%onnx::Conv_823[FLOAT, 1104x184x1x1]
%onnx::Conv_824[FLOAT, 1104]
%onnx::Conv_826[FLOAT, 1104x1x3x3]
%onnx::Conv_829[FLOAT, 184x1104x1x1]
%onnx::Conv_832[FLOAT, 1104x184x1x1]
%onnx::Conv_835[FLOAT, 1104x1x5x5]
%onnx::Conv_838[FLOAT, 184x1104x1x1]
%onnx::Conv_841[FLOAT, 184x184x1x1]
%onnx::Conv_844[FLOAT, 184x1x3x3]
%onnx::Conv_847[FLOAT, 184x184x1x1]
%onnx::Conv_850[FLOAT, 184x184x1x1]
%onnx::Conv_853[FLOAT, 184x1x3x3]
%onnx::Conv_856[FLOAT, 352x184x1x1]
%onnx::Conv_857[FLOAT, 352]
%onnx::Conv_859[FLOAT, 1504x352x1x1]
%onnx::Conv_860[FLOAT, 1504]
) {
%onnx::Conv_854 = Identity(%onnx::Conv_821)
%onnx::Conv_851 = Identity(%onnx::Conv_821)
%onnx::Conv_848 = Identity(%onnx::Conv_821)
%onnx::Conv_845 = Identity(%onnx::Conv_821)
%onnx::Conv_842 = Identity(%onnx::Conv_821)
%onnx::Conv_839 = Identity(%onnx::Conv_821)
%onnx::Conv_836 = Identity(%onnx::Conv_824)
%onnx::Conv_833 = Identity(%onnx::Conv_824)
%onnx::Conv_830 = Identity(%onnx::Conv_821)
%onnx::Conv_827 = Identity(%onnx::Conv_824)
%onnx::Conv_818 = Identity(%onnx::Conv_794)
%onnx::Conv_815 = Identity(%onnx::Conv_794)
%onnx::Conv_812 = Identity(%onnx::Conv_794)
%onnx::Conv_809 = Identity(%onnx::Conv_806)
%onnx::Conv_803 = Identity(%onnx::Conv_794)
%onnx::Conv_800 = Identity(%onnx::Conv_797)
%onnx::Conv_791 = Identity(%onnx::Conv_770)
%onnx::Conv_788 = Identity(%onnx::Conv_770)
%onnx::Conv_785 = Identity(%onnx::Conv_758)
%onnx::Conv_782 = Identity(%onnx::Conv_758)
%onnx::Conv_779 = Identity(%onnx::Conv_758)
%onnx::Conv_776 = Identity(%onnx::Conv_758)
%onnx::Conv_773 = Identity(%onnx::Conv_770)
%onnx::Conv_767 = Identity(%onnx::Conv_758)
%onnx::Conv_764 = Identity(%onnx::Conv_758)
%onnx::Conv_761 = Identity(%onnx::Conv_758)
%onnx::Conv_755 = Identity(%onnx::Conv_722)
%onnx::Conv_752 = Identity(%onnx::Conv_722)
%onnx::Conv_749 = Identity(%onnx::Conv_722)
%onnx::Conv_746 = Identity(%onnx::Conv_722)
%onnx::Conv_743 = Identity(%onnx::Conv_722)
%onnx::Conv_740 = Identity(%onnx::Conv_722)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_731 = Identity(%onnx::Conv_722)
%onnx::Conv_728 = Identity(%onnx::Conv_722)
%onnx::Conv_725 = Identity(%onnx::Conv_722)
%onnx::Conv_719 = Identity(%onnx::Conv_686)
%onnx::Conv_716 = Identity(%onnx::Conv_686)
%onnx::Conv_713 = Identity(%onnx::Conv_686)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_686)
%onnx::Conv_701 = Identity(%onnx::Conv_698)
%onnx::Conv_695 = Identity(%onnx::Conv_686)
%onnx::Conv_692 = Identity(%onnx::Conv_686)
%onnx::Conv_689 = Identity(%onnx::Conv_686)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_677 = Identity(%onnx::Conv_668)
%onnx::Conv_674 = Identity(%onnx::Conv_668)
%onnx::Conv_671 = Identity(%onnx::Conv_668)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_667, %onnx::Conv_668)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%665 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %665
}
|
val_accuracy
| 0 | 86,665,600 | 2,210,204 |
{'zcp_synflow': 80.80918186209149, 'zcp_zen': 71.32733154296875, 'zcp_epe_nas': 17.641774319685073, 'zcp_fisher': 0.1637762486934662, 'zcp_flops': 86665600.0, 'zcp_grad_norm': 29.637399673461914, 'zcp_grasp': 0.005390167236328125, 'zcp_jacov': -16.059959237778152, 'zcp_l2_norm': 669.4696044921875, 'zcp_nwot': 215.44262561877449, 'zcp_params': 2210204.0, 'zcp_plain': -0.0008670223178341985, 'zcp_snip': 45.76626205444336, 'lat_1080ti_1': 0.6733067756686864, 'lat_1080ti_32': 0.7253947824726745, 'lat_1080ti_64': 0.6920627113901126, 'lat_2080ti_1': 0.6894476376354367, 'lat_2080ti_32': 0.7117484095924766, 'lat_2080ti_64': 0.6749229934213825, 'lat_essential_ph_1': 0.49056603773584906, 'lat_eyeriss': 0.7041974584196726, 'lat_fpga': 0.6781780939847019, 'lat_gold_6226': 0.5010862804197125, 'lat_gold_6240': 0.6569627076928091, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.7535745645599249, 'lat_raspi4': 0.7236208372863555, 'lat_samsung_a50': 0.2631578947368421, 'lat_samsung_s7': 0.2992125984251969, 'lat_silver_4114': 0.6880434761133387, 'lat_silver_4210r': 0.729439519936368, 'lat_titan_rtx_1': 0.6719912150672334, 'lat_titan_rtx_32': 0.6902333226819677, 'lat_titan_rtx_64': 0.7127469195225454, 'lat_titanx_1': 0.3555717526529926, 'lat_titanx_32': 0.7417455755691775, 'lat_titanx_64': 0.6743904417288239, 'lat_titanxp_1': 0.6221004577605784, 'lat_titanxp_32': 0.7250489220134635, 'lat_titanxp_64': 0.7231337082814926}
| |
FBNet_768
|
FBNet
|
768
|
768
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_678[FLOAT, 16x3x3x3]
%onnx::Conv_679[FLOAT, 16]
%onnx::Conv_681[FLOAT, 16x8x1x1]
%onnx::Conv_684[FLOAT, 16x1x5x5]
%onnx::Conv_687[FLOAT, 16x8x1x1]
%onnx::Conv_690[FLOAT, 16x8x1x1]
%onnx::Conv_693[FLOAT, 16x1x3x3]
%onnx::Conv_696[FLOAT, 24x8x1x1]
%onnx::Conv_697[FLOAT, 24]
%onnx::Conv_699[FLOAT, 144x24x1x1]
%onnx::Conv_700[FLOAT, 144]
%onnx::Conv_702[FLOAT, 144x1x3x3]
%onnx::Conv_705[FLOAT, 24x144x1x1]
%onnx::Conv_708[FLOAT, 72x24x1x1]
%onnx::Conv_709[FLOAT, 72]
%onnx::Conv_711[FLOAT, 72x1x5x5]
%onnx::Conv_714[FLOAT, 24x72x1x1]
%onnx::Conv_717[FLOAT, 24x24x1x1]
%onnx::Conv_720[FLOAT, 24x1x5x5]
%onnx::Conv_723[FLOAT, 32x24x1x1]
%onnx::Conv_724[FLOAT, 32]
%onnx::Conv_726[FLOAT, 32x32x1x1]
%onnx::Conv_729[FLOAT, 32x1x3x3]
%onnx::Conv_732[FLOAT, 32x32x1x1]
%onnx::Conv_735[FLOAT, 96x32x1x1]
%onnx::Conv_736[FLOAT, 96]
%onnx::Conv_738[FLOAT, 96x1x5x5]
%onnx::Conv_741[FLOAT, 32x96x1x1]
%onnx::Conv_744[FLOAT, 32x16x1x1]
%onnx::Conv_747[FLOAT, 32x1x5x5]
%onnx::Conv_750[FLOAT, 32x16x1x1]
%onnx::Conv_753[FLOAT, 32x16x1x1]
%onnx::Conv_756[FLOAT, 32x1x5x5]
%onnx::Conv_759[FLOAT, 64x16x1x1]
%onnx::Conv_760[FLOAT, 64]
%onnx::Conv_762[FLOAT, 64x32x1x1]
%onnx::Conv_765[FLOAT, 64x1x3x3]
%onnx::Conv_768[FLOAT, 64x32x1x1]
%onnx::Conv_771[FLOAT, 64x64x1x1]
%onnx::Conv_774[FLOAT, 64x1x3x3]
%onnx::Conv_777[FLOAT, 64x64x1x1]
%onnx::Conv_780[FLOAT, 64x32x1x1]
%onnx::Conv_783[FLOAT, 64x1x5x5]
%onnx::Conv_786[FLOAT, 64x32x1x1]
%onnx::Conv_789[FLOAT, 384x64x1x1]
%onnx::Conv_790[FLOAT, 384]
%onnx::Conv_792[FLOAT, 384x1x3x3]
%onnx::Conv_795[FLOAT, 112x384x1x1]
%onnx::Conv_796[FLOAT, 112]
%onnx::Conv_798[FLOAT, 112x56x1x1]
%onnx::Conv_801[FLOAT, 112x1x5x5]
%onnx::Conv_804[FLOAT, 112x56x1x1]
%onnx::Conv_807[FLOAT, 112x112x1x1]
%onnx::Conv_810[FLOAT, 112x1x5x5]
%onnx::Conv_813[FLOAT, 112x112x1x1]
%onnx::Conv_816[FLOAT, 672x112x1x1]
%onnx::Conv_817[FLOAT, 672]
%onnx::Conv_819[FLOAT, 672x1x3x3]
%onnx::Conv_822[FLOAT, 184x672x1x1]
%onnx::Conv_823[FLOAT, 184]
%onnx::Conv_825[FLOAT, 552x184x1x1]
%onnx::Conv_826[FLOAT, 552]
%onnx::Conv_828[FLOAT, 552x1x5x5]
%onnx::Conv_831[FLOAT, 184x552x1x1]
%onnx::Conv_834[FLOAT, 1104x184x1x1]
%onnx::Conv_835[FLOAT, 1104]
%onnx::Conv_837[FLOAT, 1104x1x5x5]
%onnx::Conv_840[FLOAT, 184x1104x1x1]
%onnx::Conv_843[FLOAT, 184x184x1x1]
%onnx::Conv_846[FLOAT, 184x1x3x3]
%onnx::Conv_849[FLOAT, 184x184x1x1]
%onnx::Conv_852[FLOAT, 352x184x1x1]
%onnx::Conv_853[FLOAT, 352]
%onnx::Conv_855[FLOAT, 1504x352x1x1]
%onnx::Conv_856[FLOAT, 1504]
) {
%onnx::Conv_850 = Identity(%onnx::Conv_823)
%onnx::Conv_847 = Identity(%onnx::Conv_823)
%onnx::Conv_844 = Identity(%onnx::Conv_823)
%onnx::Conv_841 = Identity(%onnx::Conv_823)
%onnx::Conv_838 = Identity(%onnx::Conv_835)
%onnx::Conv_832 = Identity(%onnx::Conv_823)
%onnx::Conv_829 = Identity(%onnx::Conv_826)
%onnx::Conv_820 = Identity(%onnx::Conv_817)
%onnx::Conv_814 = Identity(%onnx::Conv_796)
%onnx::Conv_811 = Identity(%onnx::Conv_796)
%onnx::Conv_808 = Identity(%onnx::Conv_796)
%onnx::Conv_805 = Identity(%onnx::Conv_796)
%onnx::Conv_802 = Identity(%onnx::Conv_796)
%onnx::Conv_799 = Identity(%onnx::Conv_796)
%onnx::Conv_793 = Identity(%onnx::Conv_790)
%onnx::Conv_787 = Identity(%onnx::Conv_760)
%onnx::Conv_784 = Identity(%onnx::Conv_760)
%onnx::Conv_781 = Identity(%onnx::Conv_760)
%onnx::Conv_778 = Identity(%onnx::Conv_760)
%onnx::Conv_775 = Identity(%onnx::Conv_760)
%onnx::Conv_772 = Identity(%onnx::Conv_760)
%onnx::Conv_769 = Identity(%onnx::Conv_760)
%onnx::Conv_766 = Identity(%onnx::Conv_760)
%onnx::Conv_763 = Identity(%onnx::Conv_760)
%onnx::Conv_757 = Identity(%onnx::Conv_724)
%onnx::Conv_754 = Identity(%onnx::Conv_724)
%onnx::Conv_751 = Identity(%onnx::Conv_724)
%onnx::Conv_748 = Identity(%onnx::Conv_724)
%onnx::Conv_745 = Identity(%onnx::Conv_724)
%onnx::Conv_742 = Identity(%onnx::Conv_724)
%onnx::Conv_739 = Identity(%onnx::Conv_736)
%onnx::Conv_733 = Identity(%onnx::Conv_724)
%onnx::Conv_730 = Identity(%onnx::Conv_724)
%onnx::Conv_727 = Identity(%onnx::Conv_724)
%onnx::Conv_721 = Identity(%onnx::Conv_697)
%onnx::Conv_718 = Identity(%onnx::Conv_697)
%onnx::Conv_715 = Identity(%onnx::Conv_697)
%onnx::Conv_712 = Identity(%onnx::Conv_709)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_700)
%onnx::Conv_694 = Identity(%onnx::Conv_679)
%onnx::Conv_691 = Identity(%onnx::Conv_679)
%onnx::Conv_688 = Identity(%onnx::Conv_679)
%onnx::Conv_685 = Identity(%onnx::Conv_679)
%onnx::Conv_682 = Identity(%onnx::Conv_679)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_678, %onnx::Conv_679)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%676 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %676
}
|
val_accuracy
| 0 | 59,020,160 | 1,857,156 |
{'zcp_synflow': 71.84280470983647, 'zcp_zen': 62.557945251464844, 'zcp_epe_nas': 11.601880967080385, 'zcp_fisher': 0.07879749685525894, 'zcp_flops': 59020160.0, 'zcp_grad_norm': 21.606355667114258, 'zcp_grasp': 0.0004749298095703125, 'zcp_jacov': -16.068019713356797, 'zcp_l2_norm': 562.8466796875, 'zcp_nwot': 209.58425220647723, 'zcp_params': 1857156.0, 'zcp_plain': 0.005425062030553818, 'zcp_snip': 30.372377395629883, 'lat_1080ti_1': 0.5675362798681708, 'lat_1080ti_32': 0.5678913862239627, 'lat_1080ti_64': 0.382241314344138, 'lat_2080ti_1': 0.5807379361850107, 'lat_2080ti_32': 0.55966917670833, 'lat_2080ti_64': 0.4156838538814786, 'lat_essential_ph_1': 0.3584905660377358, 'lat_eyeriss': 0.3395375277354998, 'lat_fpga': 0.31444007955748615, 'lat_gold_6226': 0.26457706143099513, 'lat_gold_6240': 0.4329159301125605, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.3380308972655055, 'lat_raspi4': 0.37530354065822336, 'lat_samsung_a50': 0.14736842105263157, 'lat_samsung_s7': 0.1889763779527559, 'lat_silver_4114': 0.49567767254138867, 'lat_silver_4210r': 0.42225058899464, 'lat_titan_rtx_1': 0.5677868898954676, 'lat_titan_rtx_32': 0.5509627251948406, 'lat_titan_rtx_64': 0.45975532967769, 'lat_titanx_1': 0.3045041752719441, 'lat_titanx_32': 0.47497094875360313, 'lat_titanx_64': 0.3631461101880417, 'lat_titanxp_1': 0.5412300977173778, 'lat_titanxp_32': 0.519182240117277, 'lat_titanxp_64': 0.40951606360210346}
| |
FBNet_3531
|
FBNet
|
3531
|
3531
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_577[FLOAT, 16x3x3x3]
%onnx::Conv_578[FLOAT, 16]
%onnx::Conv_580[FLOAT, 16x16x1x1]
%onnx::Conv_583[FLOAT, 16x1x5x5]
%onnx::Conv_586[FLOAT, 16x16x1x1]
%onnx::Conv_589[FLOAT, 48x16x1x1]
%onnx::Conv_590[FLOAT, 48]
%onnx::Conv_592[FLOAT, 48x1x3x3]
%onnx::Conv_595[FLOAT, 24x48x1x1]
%onnx::Conv_596[FLOAT, 24]
%onnx::Conv_598[FLOAT, 144x24x1x1]
%onnx::Conv_599[FLOAT, 144]
%onnx::Conv_601[FLOAT, 144x1x5x5]
%onnx::Conv_604[FLOAT, 24x144x1x1]
%onnx::Conv_607[FLOAT, 24x24x1x1]
%onnx::Conv_610[FLOAT, 24x1x5x5]
%onnx::Conv_613[FLOAT, 24x24x1x1]
%onnx::Conv_616[FLOAT, 144x24x1x1]
%onnx::Conv_619[FLOAT, 144x1x3x3]
%onnx::Conv_622[FLOAT, 24x144x1x1]
%onnx::Conv_625[FLOAT, 24x24x1x1]
%onnx::Conv_628[FLOAT, 24x1x3x3]
%onnx::Conv_631[FLOAT, 32x24x1x1]
%onnx::Conv_632[FLOAT, 32]
%onnx::Conv_634[FLOAT, 192x32x1x1]
%onnx::Conv_635[FLOAT, 192]
%onnx::Conv_637[FLOAT, 192x1x3x3]
%onnx::Conv_640[FLOAT, 32x192x1x1]
%onnx::Conv_643[FLOAT, 32x16x1x1]
%onnx::Conv_646[FLOAT, 32x1x5x5]
%onnx::Conv_649[FLOAT, 32x16x1x1]
%onnx::Conv_652[FLOAT, 64x32x1x1]
%onnx::Conv_653[FLOAT, 64]
%onnx::Conv_655[FLOAT, 64x32x1x1]
%onnx::Conv_658[FLOAT, 64x1x5x5]
%onnx::Conv_661[FLOAT, 64x32x1x1]
%onnx::Conv_664[FLOAT, 64x32x1x1]
%onnx::Conv_667[FLOAT, 64x1x5x5]
%onnx::Conv_670[FLOAT, 64x32x1x1]
%onnx::Conv_673[FLOAT, 64x64x1x1]
%onnx::Conv_676[FLOAT, 64x1x5x5]
%onnx::Conv_679[FLOAT, 112x64x1x1]
%onnx::Conv_680[FLOAT, 112]
%onnx::Conv_682[FLOAT, 672x112x1x1]
%onnx::Conv_683[FLOAT, 672]
%onnx::Conv_685[FLOAT, 672x1x3x3]
%onnx::Conv_688[FLOAT, 112x672x1x1]
%onnx::Conv_691[FLOAT, 672x112x1x1]
%onnx::Conv_694[FLOAT, 672x1x3x3]
%onnx::Conv_697[FLOAT, 112x672x1x1]
%onnx::Conv_700[FLOAT, 672x112x1x1]
%onnx::Conv_703[FLOAT, 672x1x3x3]
%onnx::Conv_706[FLOAT, 184x672x1x1]
%onnx::Conv_707[FLOAT, 184]
%onnx::Conv_709[FLOAT, 1104x184x1x1]
%onnx::Conv_710[FLOAT, 1104]
%onnx::Conv_712[FLOAT, 1104x1x3x3]
%onnx::Conv_715[FLOAT, 184x1104x1x1]
%onnx::Conv_718[FLOAT, 1104x184x1x1]
%onnx::Conv_721[FLOAT, 1104x1x3x3]
%onnx::Conv_724[FLOAT, 184x1104x1x1]
%onnx::Conv_727[FLOAT, 552x184x1x1]
%onnx::Conv_728[FLOAT, 552]
%onnx::Conv_730[FLOAT, 552x1x5x5]
%onnx::Conv_733[FLOAT, 184x552x1x1]
%onnx::Conv_736[FLOAT, 184x184x1x1]
%onnx::Conv_739[FLOAT, 184x1x3x3]
%onnx::Conv_742[FLOAT, 352x184x1x1]
%onnx::Conv_743[FLOAT, 352]
%onnx::Conv_745[FLOAT, 1504x352x1x1]
%onnx::Conv_746[FLOAT, 1504]
) {
%onnx::Conv_740 = Identity(%onnx::Conv_707)
%onnx::Conv_737 = Identity(%onnx::Conv_707)
%onnx::Conv_734 = Identity(%onnx::Conv_707)
%onnx::Conv_731 = Identity(%onnx::Conv_728)
%onnx::Conv_725 = Identity(%onnx::Conv_707)
%onnx::Conv_722 = Identity(%onnx::Conv_710)
%onnx::Conv_719 = Identity(%onnx::Conv_710)
%onnx::Conv_716 = Identity(%onnx::Conv_707)
%onnx::Conv_713 = Identity(%onnx::Conv_710)
%onnx::Conv_704 = Identity(%onnx::Conv_683)
%onnx::Conv_701 = Identity(%onnx::Conv_683)
%onnx::Conv_698 = Identity(%onnx::Conv_680)
%onnx::Conv_695 = Identity(%onnx::Conv_683)
%onnx::Conv_692 = Identity(%onnx::Conv_683)
%onnx::Conv_689 = Identity(%onnx::Conv_680)
%onnx::Conv_686 = Identity(%onnx::Conv_683)
%onnx::Conv_677 = Identity(%onnx::Conv_653)
%onnx::Conv_674 = Identity(%onnx::Conv_653)
%onnx::Conv_671 = Identity(%onnx::Conv_653)
%onnx::Conv_668 = Identity(%onnx::Conv_653)
%onnx::Conv_665 = Identity(%onnx::Conv_653)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_650 = Identity(%onnx::Conv_632)
%onnx::Conv_647 = Identity(%onnx::Conv_632)
%onnx::Conv_644 = Identity(%onnx::Conv_632)
%onnx::Conv_641 = Identity(%onnx::Conv_632)
%onnx::Conv_638 = Identity(%onnx::Conv_635)
%onnx::Conv_629 = Identity(%onnx::Conv_596)
%onnx::Conv_626 = Identity(%onnx::Conv_596)
%onnx::Conv_623 = Identity(%onnx::Conv_596)
%onnx::Conv_620 = Identity(%onnx::Conv_599)
%onnx::Conv_617 = Identity(%onnx::Conv_599)
%onnx::Conv_614 = Identity(%onnx::Conv_596)
%onnx::Conv_611 = Identity(%onnx::Conv_596)
%onnx::Conv_608 = Identity(%onnx::Conv_596)
%onnx::Conv_605 = Identity(%onnx::Conv_596)
%onnx::Conv_602 = Identity(%onnx::Conv_599)
%onnx::Conv_593 = Identity(%onnx::Conv_590)
%onnx::Conv_587 = Identity(%onnx::Conv_578)
%onnx::Conv_584 = Identity(%onnx::Conv_578)
%onnx::Conv_581 = Identity(%onnx::Conv_578)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_577, %onnx::Conv_578)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_580, %onnx::Conv_581)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_583, %onnx::Conv_584)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_586, %onnx::Conv_587)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_589, %onnx::Conv_590)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_592, %onnx::Conv_593)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_595, %onnx::Conv_596)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_598, %onnx::Conv_599)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_601, %onnx::Conv_602)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_604, %onnx::Conv_605)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_607, %onnx::Conv_608)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_610, %onnx::Conv_611)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_613, %onnx::Conv_614)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_616, %onnx::Conv_617)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_619, %onnx::Conv_620)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_622, %onnx::Conv_623)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_625, %onnx::Conv_626)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_628, %onnx::Conv_629)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_631, %onnx::Conv_632)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%575 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %575
}
|
val_accuracy
| 0 | 88,932,224 | 2,445,316 |
{'zcp_synflow': 72.95692860811248, 'zcp_zen': 64.07454681396484, 'zcp_epe_nas': 10.028392449947253, 'zcp_fisher': 0.10712198168039322, 'zcp_flops': 88932224.0, 'zcp_grad_norm': 20.186874389648438, 'zcp_grasp': -0.07990169525146484, 'zcp_jacov': -16.054545784770895, 'zcp_l2_norm': 637.5703735351562, 'zcp_nwot': 215.67261300572991, 'zcp_params': 2445316.0, 'zcp_plain': -0.00019556637562345713, 'zcp_snip': 40.82132339477539, 'lat_1080ti_1': 0.41789725620165924, 'lat_1080ti_32': 0.4813234941424713, 'lat_1080ti_64': 0.5320264981630543, 'lat_2080ti_1': 0.4324272968502251, 'lat_2080ti_32': 0.4965372329761316, 'lat_2080ti_64': 0.5272402858112675, 'lat_essential_ph_1': 0.4339622641509434, 'lat_eyeriss': 0.6407863128747732, 'lat_fpga': 0.740666343423149, 'lat_gold_6226': 0.5205321452296754, 'lat_gold_6240': 0.6411784636682651, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.6188932995280692, 'lat_raspi4': 0.6718541154302439, 'lat_samsung_a50': 0.2631578947368421, 'lat_samsung_s7': 0.2283464566929134, 'lat_silver_4114': 0.6993986450372478, 'lat_silver_4210r': 0.6431164811895596, 'lat_titan_rtx_1': 0.39811504358821054, 'lat_titan_rtx_32': 0.45430766346562, 'lat_titan_rtx_64': 0.5031429680964549, 'lat_titanx_1': 0.22988790873408954, 'lat_titanx_32': 0.47903493440381867, 'lat_titanx_64': 0.5046708148221388, 'lat_titanxp_1': 0.4011317374921298, 'lat_titanxp_32': 0.47993468993861416, 'lat_titanxp_64': 0.5207146777875797}
| |
FBNet_3409
|
FBNet
|
3409
|
3409
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_696[FLOAT, 16x3x3x3]
%onnx::Conv_697[FLOAT, 16]
%onnx::Conv_699[FLOAT, 16x8x1x1]
%onnx::Conv_702[FLOAT, 16x1x5x5]
%onnx::Conv_705[FLOAT, 16x8x1x1]
%onnx::Conv_708[FLOAT, 16x8x1x1]
%onnx::Conv_711[FLOAT, 16x1x5x5]
%onnx::Conv_714[FLOAT, 24x8x1x1]
%onnx::Conv_715[FLOAT, 24]
%onnx::Conv_717[FLOAT, 24x12x1x1]
%onnx::Conv_720[FLOAT, 24x1x5x5]
%onnx::Conv_723[FLOAT, 24x12x1x1]
%onnx::Conv_726[FLOAT, 72x24x1x1]
%onnx::Conv_727[FLOAT, 72]
%onnx::Conv_729[FLOAT, 72x1x3x3]
%onnx::Conv_732[FLOAT, 24x72x1x1]
%onnx::Conv_735[FLOAT, 24x24x1x1]
%onnx::Conv_738[FLOAT, 24x1x3x3]
%onnx::Conv_741[FLOAT, 24x24x1x1]
%onnx::Conv_744[FLOAT, 144x24x1x1]
%onnx::Conv_745[FLOAT, 144]
%onnx::Conv_747[FLOAT, 144x1x3x3]
%onnx::Conv_750[FLOAT, 32x144x1x1]
%onnx::Conv_751[FLOAT, 32]
%onnx::Conv_753[FLOAT, 192x32x1x1]
%onnx::Conv_754[FLOAT, 192]
%onnx::Conv_756[FLOAT, 192x1x5x5]
%onnx::Conv_759[FLOAT, 32x192x1x1]
%onnx::Conv_762[FLOAT, 192x32x1x1]
%onnx::Conv_765[FLOAT, 192x1x3x3]
%onnx::Conv_768[FLOAT, 32x192x1x1]
%onnx::Conv_771[FLOAT, 32x16x1x1]
%onnx::Conv_774[FLOAT, 32x1x5x5]
%onnx::Conv_777[FLOAT, 32x16x1x1]
%onnx::Conv_780[FLOAT, 64x32x1x1]
%onnx::Conv_781[FLOAT, 64]
%onnx::Conv_783[FLOAT, 192x64x1x1]
%onnx::Conv_786[FLOAT, 192x1x5x5]
%onnx::Conv_789[FLOAT, 64x192x1x1]
%onnx::Conv_792[FLOAT, 384x64x1x1]
%onnx::Conv_793[FLOAT, 384]
%onnx::Conv_795[FLOAT, 384x1x3x3]
%onnx::Conv_798[FLOAT, 64x384x1x1]
%onnx::Conv_801[FLOAT, 64x64x1x1]
%onnx::Conv_804[FLOAT, 64x1x5x5]
%onnx::Conv_807[FLOAT, 64x64x1x1]
%onnx::Conv_810[FLOAT, 64x64x1x1]
%onnx::Conv_813[FLOAT, 64x1x5x5]
%onnx::Conv_816[FLOAT, 112x64x1x1]
%onnx::Conv_817[FLOAT, 112]
%onnx::Conv_819[FLOAT, 112x56x1x1]
%onnx::Conv_822[FLOAT, 112x1x3x3]
%onnx::Conv_825[FLOAT, 112x56x1x1]
%onnx::Conv_828[FLOAT, 112x112x1x1]
%onnx::Conv_831[FLOAT, 112x1x3x3]
%onnx::Conv_834[FLOAT, 112x112x1x1]
%onnx::Conv_837[FLOAT, 672x112x1x1]
%onnx::Conv_838[FLOAT, 672]
%onnx::Conv_840[FLOAT, 672x1x3x3]
%onnx::Conv_843[FLOAT, 112x672x1x1]
%onnx::Conv_846[FLOAT, 672x112x1x1]
%onnx::Conv_849[FLOAT, 672x1x3x3]
%onnx::Conv_852[FLOAT, 184x672x1x1]
%onnx::Conv_853[FLOAT, 184]
%onnx::Conv_855[FLOAT, 552x184x1x1]
%onnx::Conv_856[FLOAT, 552]
%onnx::Conv_858[FLOAT, 552x1x5x5]
%onnx::Conv_861[FLOAT, 184x552x1x1]
%onnx::Conv_864[FLOAT, 1104x184x1x1]
%onnx::Conv_865[FLOAT, 1104]
%onnx::Conv_867[FLOAT, 1104x1x5x5]
%onnx::Conv_870[FLOAT, 184x1104x1x1]
%onnx::Conv_873[FLOAT, 184x184x1x1]
%onnx::Conv_876[FLOAT, 184x1x3x3]
%onnx::Conv_879[FLOAT, 184x184x1x1]
%onnx::Conv_882[FLOAT, 1104x184x1x1]
%onnx::Conv_885[FLOAT, 1104x1x3x3]
%onnx::Conv_888[FLOAT, 352x1104x1x1]
%onnx::Conv_889[FLOAT, 352]
%onnx::Conv_891[FLOAT, 1504x352x1x1]
%onnx::Conv_892[FLOAT, 1504]
) {
%onnx::Conv_886 = Identity(%onnx::Conv_865)
%onnx::Conv_883 = Identity(%onnx::Conv_865)
%onnx::Conv_880 = Identity(%onnx::Conv_853)
%onnx::Conv_877 = Identity(%onnx::Conv_853)
%onnx::Conv_874 = Identity(%onnx::Conv_853)
%onnx::Conv_871 = Identity(%onnx::Conv_853)
%onnx::Conv_868 = Identity(%onnx::Conv_865)
%onnx::Conv_862 = Identity(%onnx::Conv_853)
%onnx::Conv_859 = Identity(%onnx::Conv_856)
%onnx::Conv_850 = Identity(%onnx::Conv_838)
%onnx::Conv_847 = Identity(%onnx::Conv_838)
%onnx::Conv_844 = Identity(%onnx::Conv_817)
%onnx::Conv_841 = Identity(%onnx::Conv_838)
%onnx::Conv_835 = Identity(%onnx::Conv_817)
%onnx::Conv_832 = Identity(%onnx::Conv_817)
%onnx::Conv_829 = Identity(%onnx::Conv_817)
%onnx::Conv_826 = Identity(%onnx::Conv_817)
%onnx::Conv_823 = Identity(%onnx::Conv_817)
%onnx::Conv_820 = Identity(%onnx::Conv_817)
%onnx::Conv_814 = Identity(%onnx::Conv_781)
%onnx::Conv_811 = Identity(%onnx::Conv_781)
%onnx::Conv_808 = Identity(%onnx::Conv_781)
%onnx::Conv_805 = Identity(%onnx::Conv_781)
%onnx::Conv_802 = Identity(%onnx::Conv_781)
%onnx::Conv_799 = Identity(%onnx::Conv_781)
%onnx::Conv_796 = Identity(%onnx::Conv_793)
%onnx::Conv_790 = Identity(%onnx::Conv_781)
%onnx::Conv_787 = Identity(%onnx::Conv_754)
%onnx::Conv_784 = Identity(%onnx::Conv_754)
%onnx::Conv_778 = Identity(%onnx::Conv_751)
%onnx::Conv_775 = Identity(%onnx::Conv_751)
%onnx::Conv_772 = Identity(%onnx::Conv_751)
%onnx::Conv_769 = Identity(%onnx::Conv_751)
%onnx::Conv_766 = Identity(%onnx::Conv_754)
%onnx::Conv_763 = Identity(%onnx::Conv_754)
%onnx::Conv_760 = Identity(%onnx::Conv_751)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_748 = Identity(%onnx::Conv_745)
%onnx::Conv_742 = Identity(%onnx::Conv_715)
%onnx::Conv_739 = Identity(%onnx::Conv_715)
%onnx::Conv_736 = Identity(%onnx::Conv_715)
%onnx::Conv_733 = Identity(%onnx::Conv_715)
%onnx::Conv_730 = Identity(%onnx::Conv_727)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_715)
%onnx::Conv_718 = Identity(%onnx::Conv_715)
%onnx::Conv_712 = Identity(%onnx::Conv_697)
%onnx::Conv_709 = Identity(%onnx::Conv_697)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_697)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_696, %onnx::Conv_697)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.8/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.9/relu/Relu_output_0 = Relu(%/cells.9/conv/Conv_output_0)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/relu/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/relu/Relu_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_882, %onnx::Conv_883)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_885, %onnx::Conv_886)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_888, %onnx::Conv_889)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_891, %onnx::Conv_892)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%694 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %694
}
|
val_accuracy
| 0 | 81,339,264 | 2,591,196 |
{'zcp_synflow': 81.14917440093608, 'zcp_zen': 73.0303955078125, 'zcp_epe_nas': 15.927021549794286, 'zcp_fisher': 0.1390276700258255, 'zcp_flops': 81339264.0, 'zcp_grad_norm': 28.084260940551758, 'zcp_grasp': -0.001708984375, 'zcp_jacov': -16.062695122181076, 'zcp_l2_norm': 708.3450317382812, 'zcp_nwot': 213.06082189908167, 'zcp_params': 2591196.0, 'zcp_plain': -0.0034320573322474957, 'zcp_snip': 46.48149490356445, 'lat_1080ti_1': 0.790860868124381, 'lat_1080ti_32': 0.541912793987485, 'lat_1080ti_64': 0.4612260381484114, 'lat_2080ti_1': 0.7705408862696465, 'lat_2080ti_32': 0.6113131664622383, 'lat_2080ti_64': 0.5030559399218113, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.5799515889461428, 'lat_fpga': 0.6325314915049624, 'lat_gold_6226': 0.5308825796242859, 'lat_gold_6240': 0.743111924616377, 'lat_pixel2': 0.5217391304347826, 'lat_pixel3': 0.543193261383369, 'lat_raspi4': 0.6249245179154561, 'lat_samsung_a50': 0.28421052631578947, 'lat_samsung_s7': 0.6614173228346457, 'lat_silver_4114': 0.7596685652022273, 'lat_silver_4210r': 0.7806125252800129, 'lat_titan_rtx_1': 0.7222847972989977, 'lat_titan_rtx_32': 0.5942416934566236, 'lat_titan_rtx_64': 0.5207455019293918, 'lat_titanx_1': 0.39082173540219134, 'lat_titanx_32': 0.5184801335778153, 'lat_titanx_64': 0.4916918946091258, 'lat_titanxp_1': 0.7013579040492496, 'lat_titanxp_32': 0.5584574818365604, 'lat_titanxp_64': 0.459872135528553}
| |
FBNet_2926
|
FBNet
|
2926
|
2926
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_686[FLOAT, 16x3x3x3]
%onnx::Conv_687[FLOAT, 16]
%onnx::Conv_689[FLOAT, 48x16x1x1]
%onnx::Conv_690[FLOAT, 48]
%onnx::Conv_692[FLOAT, 48x1x5x5]
%onnx::Conv_695[FLOAT, 24x48x1x1]
%onnx::Conv_696[FLOAT, 24]
%onnx::Conv_698[FLOAT, 24x24x1x1]
%onnx::Conv_701[FLOAT, 24x1x3x3]
%onnx::Conv_704[FLOAT, 24x24x1x1]
%onnx::Conv_707[FLOAT, 72x24x1x1]
%onnx::Conv_708[FLOAT, 72]
%onnx::Conv_710[FLOAT, 72x1x3x3]
%onnx::Conv_713[FLOAT, 32x72x1x1]
%onnx::Conv_714[FLOAT, 32]
%onnx::Conv_716[FLOAT, 96x32x1x1]
%onnx::Conv_717[FLOAT, 96]
%onnx::Conv_719[FLOAT, 96x1x5x5]
%onnx::Conv_722[FLOAT, 32x96x1x1]
%onnx::Conv_725[FLOAT, 32x16x1x1]
%onnx::Conv_728[FLOAT, 32x1x3x3]
%onnx::Conv_731[FLOAT, 32x16x1x1]
%onnx::Conv_734[FLOAT, 32x16x1x1]
%onnx::Conv_737[FLOAT, 32x1x5x5]
%onnx::Conv_740[FLOAT, 32x16x1x1]
%onnx::Conv_743[FLOAT, 32x16x1x1]
%onnx::Conv_746[FLOAT, 32x1x5x5]
%onnx::Conv_749[FLOAT, 64x16x1x1]
%onnx::Conv_750[FLOAT, 64]
%onnx::Conv_752[FLOAT, 192x64x1x1]
%onnx::Conv_753[FLOAT, 192]
%onnx::Conv_755[FLOAT, 192x1x3x3]
%onnx::Conv_758[FLOAT, 64x192x1x1]
%onnx::Conv_761[FLOAT, 64x32x1x1]
%onnx::Conv_764[FLOAT, 64x1x5x5]
%onnx::Conv_767[FLOAT, 64x32x1x1]
%onnx::Conv_770[FLOAT, 64x64x1x1]
%onnx::Conv_773[FLOAT, 64x1x5x5]
%onnx::Conv_776[FLOAT, 64x64x1x1]
%onnx::Conv_779[FLOAT, 64x32x1x1]
%onnx::Conv_782[FLOAT, 64x1x5x5]
%onnx::Conv_785[FLOAT, 112x32x1x1]
%onnx::Conv_786[FLOAT, 112]
%onnx::Conv_788[FLOAT, 112x112x1x1]
%onnx::Conv_791[FLOAT, 112x1x5x5]
%onnx::Conv_794[FLOAT, 112x112x1x1]
%onnx::Conv_797[FLOAT, 336x112x1x1]
%onnx::Conv_798[FLOAT, 336]
%onnx::Conv_800[FLOAT, 336x1x5x5]
%onnx::Conv_803[FLOAT, 112x336x1x1]
%onnx::Conv_806[FLOAT, 112x112x1x1]
%onnx::Conv_809[FLOAT, 112x1x5x5]
%onnx::Conv_812[FLOAT, 112x112x1x1]
%onnx::Conv_815[FLOAT, 112x56x1x1]
%onnx::Conv_818[FLOAT, 112x1x3x3]
%onnx::Conv_821[FLOAT, 184x56x1x1]
%onnx::Conv_822[FLOAT, 184]
%onnx::Conv_824[FLOAT, 184x92x1x1]
%onnx::Conv_827[FLOAT, 184x1x3x3]
%onnx::Conv_830[FLOAT, 184x92x1x1]
%onnx::Conv_833[FLOAT, 552x184x1x1]
%onnx::Conv_834[FLOAT, 552]
%onnx::Conv_836[FLOAT, 552x1x5x5]
%onnx::Conv_839[FLOAT, 184x552x1x1]
%onnx::Conv_842[FLOAT, 1104x184x1x1]
%onnx::Conv_843[FLOAT, 1104]
%onnx::Conv_845[FLOAT, 1104x1x5x5]
%onnx::Conv_848[FLOAT, 184x1104x1x1]
%onnx::Conv_851[FLOAT, 184x92x1x1]
%onnx::Conv_854[FLOAT, 184x1x5x5]
%onnx::Conv_857[FLOAT, 352x92x1x1]
%onnx::Conv_858[FLOAT, 352]
%onnx::Conv_860[FLOAT, 1504x352x1x1]
%onnx::Conv_861[FLOAT, 1504]
) {
%onnx::Conv_855 = Identity(%onnx::Conv_822)
%onnx::Conv_852 = Identity(%onnx::Conv_822)
%onnx::Conv_849 = Identity(%onnx::Conv_822)
%onnx::Conv_846 = Identity(%onnx::Conv_843)
%onnx::Conv_840 = Identity(%onnx::Conv_822)
%onnx::Conv_837 = Identity(%onnx::Conv_834)
%onnx::Conv_831 = Identity(%onnx::Conv_822)
%onnx::Conv_828 = Identity(%onnx::Conv_822)
%onnx::Conv_825 = Identity(%onnx::Conv_822)
%onnx::Conv_819 = Identity(%onnx::Conv_786)
%onnx::Conv_816 = Identity(%onnx::Conv_786)
%onnx::Conv_813 = Identity(%onnx::Conv_786)
%onnx::Conv_810 = Identity(%onnx::Conv_786)
%onnx::Conv_807 = Identity(%onnx::Conv_786)
%onnx::Conv_804 = Identity(%onnx::Conv_786)
%onnx::Conv_801 = Identity(%onnx::Conv_798)
%onnx::Conv_795 = Identity(%onnx::Conv_786)
%onnx::Conv_792 = Identity(%onnx::Conv_786)
%onnx::Conv_789 = Identity(%onnx::Conv_786)
%onnx::Conv_783 = Identity(%onnx::Conv_750)
%onnx::Conv_780 = Identity(%onnx::Conv_750)
%onnx::Conv_777 = Identity(%onnx::Conv_750)
%onnx::Conv_774 = Identity(%onnx::Conv_750)
%onnx::Conv_771 = Identity(%onnx::Conv_750)
%onnx::Conv_768 = Identity(%onnx::Conv_750)
%onnx::Conv_765 = Identity(%onnx::Conv_750)
%onnx::Conv_762 = Identity(%onnx::Conv_750)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_747 = Identity(%onnx::Conv_714)
%onnx::Conv_744 = Identity(%onnx::Conv_714)
%onnx::Conv_741 = Identity(%onnx::Conv_714)
%onnx::Conv_738 = Identity(%onnx::Conv_714)
%onnx::Conv_735 = Identity(%onnx::Conv_714)
%onnx::Conv_732 = Identity(%onnx::Conv_714)
%onnx::Conv_729 = Identity(%onnx::Conv_714)
%onnx::Conv_726 = Identity(%onnx::Conv_714)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%onnx::Conv_711 = Identity(%onnx::Conv_708)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_696)
%onnx::Conv_699 = Identity(%onnx::Conv_696)
%onnx::Conv_693 = Identity(%onnx::Conv_690)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_686, %onnx::Conv_687)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%684 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %684
}
|
val_accuracy
| 0 | 45,626,496 | 1,670,444 |
{'zcp_synflow': 69.9167833975955, 'zcp_zen': 62.90117263793945, 'zcp_epe_nas': 23.68429481126309, 'zcp_fisher': 0.05859559029340744, 'zcp_flops': 45626496.0, 'zcp_grad_norm': 18.236661911010742, 'zcp_grasp': 0.017566680908203125, 'zcp_jacov': -16.05602193461055, 'zcp_l2_norm': 541.0162963867188, 'zcp_nwot': 202.29422046687458, 'zcp_params': 1670444.0, 'zcp_plain': 0.005908805411309004, 'zcp_snip': 30.297626495361328, 'lat_1080ti_1': 0.6339846493408406, 'lat_1080ti_32': 0.49508775672158734, 'lat_1080ti_64': 0.20629962764836382, 'lat_2080ti_1': 0.589809381649959, 'lat_2080ti_32': 0.4387757880917416, 'lat_2080ti_64': 0.24811940350399753, 'lat_essential_ph_1': 0.18867924528301888, 'lat_eyeriss': 0.16251627454935547, 'lat_fpga': 0.18034019731048195, 'lat_gold_6226': 0.21313808201657122, 'lat_gold_6240': 0.3084430618263471, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.16389788338927475, 'lat_raspi4': 0.19339073178173483, 'lat_samsung_a50': 0.09473684210526316, 'lat_samsung_s7': 0.10236220472440945, 'lat_silver_4114': 0.3620948611614548, 'lat_silver_4210r': 0.3023158738931948, 'lat_titan_rtx_1': 0.5396307332991132, 'lat_titan_rtx_32': 0.44971910369840257, 'lat_titan_rtx_64': 0.2812853041586115, 'lat_titanx_1': 0.28933760174575857, 'lat_titanx_32': 0.3577073630143584, 'lat_titanx_64': 0.2024605427540566, 'lat_titanxp_1': 0.5187757330520784, 'lat_titanxp_32': 0.40419753439002853, 'lat_titanxp_64': 0.23945470407038819}
| |
FBNet_4197
|
FBNet
|
4197
|
4197
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_670[FLOAT, 16x3x3x3]
%onnx::Conv_671[FLOAT, 16]
%onnx::Conv_673[FLOAT, 16x8x1x1]
%onnx::Conv_676[FLOAT, 16x1x5x5]
%onnx::Conv_679[FLOAT, 16x8x1x1]
%onnx::Conv_682[FLOAT, 96x16x1x1]
%onnx::Conv_683[FLOAT, 96]
%onnx::Conv_685[FLOAT, 96x1x5x5]
%onnx::Conv_688[FLOAT, 24x96x1x1]
%onnx::Conv_689[FLOAT, 24]
%onnx::Conv_691[FLOAT, 24x12x1x1]
%onnx::Conv_694[FLOAT, 24x1x3x3]
%onnx::Conv_697[FLOAT, 24x12x1x1]
%onnx::Conv_700[FLOAT, 24x24x1x1]
%onnx::Conv_703[FLOAT, 24x1x5x5]
%onnx::Conv_706[FLOAT, 24x24x1x1]
%onnx::Conv_709[FLOAT, 24x24x1x1]
%onnx::Conv_712[FLOAT, 24x1x3x3]
%onnx::Conv_715[FLOAT, 24x24x1x1]
%onnx::Conv_718[FLOAT, 72x24x1x1]
%onnx::Conv_719[FLOAT, 72]
%onnx::Conv_721[FLOAT, 72x1x5x5]
%onnx::Conv_724[FLOAT, 32x72x1x1]
%onnx::Conv_725[FLOAT, 32]
%onnx::Conv_727[FLOAT, 96x32x1x1]
%onnx::Conv_730[FLOAT, 96x1x5x5]
%onnx::Conv_733[FLOAT, 32x96x1x1]
%onnx::Conv_736[FLOAT, 96x32x1x1]
%onnx::Conv_739[FLOAT, 96x1x5x5]
%onnx::Conv_742[FLOAT, 32x96x1x1]
%onnx::Conv_745[FLOAT, 32x16x1x1]
%onnx::Conv_748[FLOAT, 32x1x3x3]
%onnx::Conv_751[FLOAT, 64x16x1x1]
%onnx::Conv_752[FLOAT, 64]
%onnx::Conv_754[FLOAT, 192x64x1x1]
%onnx::Conv_755[FLOAT, 192]
%onnx::Conv_757[FLOAT, 192x1x3x3]
%onnx::Conv_760[FLOAT, 64x192x1x1]
%onnx::Conv_763[FLOAT, 192x64x1x1]
%onnx::Conv_766[FLOAT, 192x1x3x3]
%onnx::Conv_769[FLOAT, 64x192x1x1]
%onnx::Conv_772[FLOAT, 192x64x1x1]
%onnx::Conv_775[FLOAT, 192x1x3x3]
%onnx::Conv_778[FLOAT, 64x192x1x1]
%onnx::Conv_781[FLOAT, 112x64x1x1]
%onnx::Conv_782[FLOAT, 112]
%onnx::Conv_784[FLOAT, 336x112x1x1]
%onnx::Conv_785[FLOAT, 336]
%onnx::Conv_787[FLOAT, 336x1x3x3]
%onnx::Conv_790[FLOAT, 112x336x1x1]
%onnx::Conv_793[FLOAT, 112x56x1x1]
%onnx::Conv_796[FLOAT, 112x1x3x3]
%onnx::Conv_799[FLOAT, 112x56x1x1]
%onnx::Conv_802[FLOAT, 672x112x1x1]
%onnx::Conv_803[FLOAT, 672]
%onnx::Conv_805[FLOAT, 672x1x3x3]
%onnx::Conv_808[FLOAT, 112x672x1x1]
%onnx::Conv_811[FLOAT, 112x112x1x1]
%onnx::Conv_814[FLOAT, 112x1x5x5]
%onnx::Conv_817[FLOAT, 184x112x1x1]
%onnx::Conv_818[FLOAT, 184]
%onnx::Conv_820[FLOAT, 184x184x1x1]
%onnx::Conv_823[FLOAT, 184x1x3x3]
%onnx::Conv_826[FLOAT, 184x184x1x1]
%onnx::Conv_829[FLOAT, 552x184x1x1]
%onnx::Conv_830[FLOAT, 552]
%onnx::Conv_832[FLOAT, 552x1x5x5]
%onnx::Conv_835[FLOAT, 184x552x1x1]
%onnx::Conv_838[FLOAT, 1104x184x1x1]
%onnx::Conv_839[FLOAT, 1104]
%onnx::Conv_841[FLOAT, 1104x1x5x5]
%onnx::Conv_844[FLOAT, 184x1104x1x1]
%onnx::Conv_847[FLOAT, 184x92x1x1]
%onnx::Conv_850[FLOAT, 184x1x3x3]
%onnx::Conv_853[FLOAT, 352x92x1x1]
%onnx::Conv_854[FLOAT, 352]
%onnx::Conv_856[FLOAT, 1504x352x1x1]
%onnx::Conv_857[FLOAT, 1504]
) {
%onnx::Conv_851 = Identity(%onnx::Conv_818)
%onnx::Conv_848 = Identity(%onnx::Conv_818)
%onnx::Conv_845 = Identity(%onnx::Conv_818)
%onnx::Conv_842 = Identity(%onnx::Conv_839)
%onnx::Conv_836 = Identity(%onnx::Conv_818)
%onnx::Conv_833 = Identity(%onnx::Conv_830)
%onnx::Conv_827 = Identity(%onnx::Conv_818)
%onnx::Conv_824 = Identity(%onnx::Conv_818)
%onnx::Conv_821 = Identity(%onnx::Conv_818)
%onnx::Conv_815 = Identity(%onnx::Conv_782)
%onnx::Conv_812 = Identity(%onnx::Conv_782)
%onnx::Conv_809 = Identity(%onnx::Conv_782)
%onnx::Conv_806 = Identity(%onnx::Conv_803)
%onnx::Conv_800 = Identity(%onnx::Conv_782)
%onnx::Conv_797 = Identity(%onnx::Conv_782)
%onnx::Conv_794 = Identity(%onnx::Conv_782)
%onnx::Conv_791 = Identity(%onnx::Conv_782)
%onnx::Conv_788 = Identity(%onnx::Conv_785)
%onnx::Conv_779 = Identity(%onnx::Conv_752)
%onnx::Conv_776 = Identity(%onnx::Conv_755)
%onnx::Conv_773 = Identity(%onnx::Conv_755)
%onnx::Conv_770 = Identity(%onnx::Conv_752)
%onnx::Conv_767 = Identity(%onnx::Conv_755)
%onnx::Conv_764 = Identity(%onnx::Conv_755)
%onnx::Conv_761 = Identity(%onnx::Conv_752)
%onnx::Conv_758 = Identity(%onnx::Conv_755)
%onnx::Conv_749 = Identity(%onnx::Conv_725)
%onnx::Conv_746 = Identity(%onnx::Conv_725)
%onnx::Conv_743 = Identity(%onnx::Conv_725)
%onnx::Conv_740 = Identity(%onnx::Conv_683)
%onnx::Conv_737 = Identity(%onnx::Conv_683)
%onnx::Conv_734 = Identity(%onnx::Conv_725)
%onnx::Conv_731 = Identity(%onnx::Conv_683)
%onnx::Conv_728 = Identity(%onnx::Conv_683)
%onnx::Conv_722 = Identity(%onnx::Conv_719)
%onnx::Conv_716 = Identity(%onnx::Conv_689)
%onnx::Conv_713 = Identity(%onnx::Conv_689)
%onnx::Conv_710 = Identity(%onnx::Conv_689)
%onnx::Conv_707 = Identity(%onnx::Conv_689)
%onnx::Conv_704 = Identity(%onnx::Conv_689)
%onnx::Conv_701 = Identity(%onnx::Conv_689)
%onnx::Conv_698 = Identity(%onnx::Conv_689)
%onnx::Conv_695 = Identity(%onnx::Conv_689)
%onnx::Conv_692 = Identity(%onnx::Conv_689)
%onnx::Conv_686 = Identity(%onnx::Conv_683)
%onnx::Conv_680 = Identity(%onnx::Conv_671)
%onnx::Conv_677 = Identity(%onnx::Conv_671)
%onnx::Conv_674 = Identity(%onnx::Conv_671)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_670, %onnx::Conv_671)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%668 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %668
}
|
val_accuracy
| 0 | 65,157,248 | 1,881,772 |
{'zcp_synflow': 77.56999209851331, 'zcp_zen': 68.25639343261719, 'zcp_epe_nas': 18.775763836968423, 'zcp_fisher': 0.12732712924480438, 'zcp_flops': 65157248.0, 'zcp_grad_norm': 26.11075210571289, 'zcp_grasp': -0.02881622314453125, 'zcp_jacov': -16.05192657310107, 'zcp_l2_norm': 628.2941284179688, 'zcp_nwot': 210.10152456463234, 'zcp_params': 1881772.0, 'zcp_plain': 0.007796869613230228, 'zcp_snip': 51.899539947509766, 'lat_1080ti_1': 0.59389671553577, 'lat_1080ti_32': 0.5129523343711737, 'lat_1080ti_64': 0.3919619332342124, 'lat_2080ti_1': 0.6777511261672972, 'lat_2080ti_32': 0.5582108696854562, 'lat_2080ti_64': 0.3988704978354231, 'lat_essential_ph_1': 0.3018867924528302, 'lat_eyeriss': 0.39505437074798744, 'lat_fpga': 0.42933335972055925, 'lat_gold_6226': 0.36586053477177666, 'lat_gold_6240': 0.558921012260675, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.40174129779819817, 'lat_raspi4': 0.4419154980455869, 'lat_samsung_a50': 0.18947368421052632, 'lat_samsung_s7': 0.16535433070866143, 'lat_silver_4114': 0.5963986604365414, 'lat_silver_4210r': 0.6179418503523428, 'lat_titan_rtx_1': 0.6300110130246868, 'lat_titan_rtx_32': 0.5562317600719596, 'lat_titan_rtx_64': 0.41975664812692925, 'lat_titanx_1': 0.3308600640240046, 'lat_titanx_32': 0.4665170666502464, 'lat_titanx_64': 0.3990526264866615, 'lat_titanxp_1': 0.6002176853155193, 'lat_titanxp_32': 0.514821101209263, 'lat_titanxp_64': 0.40573829143606843}
| |
FBNet_624
|
FBNet
|
624
|
624
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_651[FLOAT, 16x3x3x3]
%onnx::Conv_652[FLOAT, 16]
%onnx::Conv_654[FLOAT, 96x16x1x1]
%onnx::Conv_655[FLOAT, 96]
%onnx::Conv_657[FLOAT, 96x1x3x3]
%onnx::Conv_660[FLOAT, 16x96x1x1]
%onnx::Conv_663[FLOAT, 16x8x1x1]
%onnx::Conv_666[FLOAT, 16x1x3x3]
%onnx::Conv_669[FLOAT, 24x8x1x1]
%onnx::Conv_670[FLOAT, 24]
%onnx::Conv_672[FLOAT, 24x24x1x1]
%onnx::Conv_675[FLOAT, 24x1x3x3]
%onnx::Conv_678[FLOAT, 24x24x1x1]
%onnx::Conv_681[FLOAT, 24x24x1x1]
%onnx::Conv_684[FLOAT, 24x1x5x5]
%onnx::Conv_687[FLOAT, 24x24x1x1]
%onnx::Conv_690[FLOAT, 72x24x1x1]
%onnx::Conv_691[FLOAT, 72]
%onnx::Conv_693[FLOAT, 72x1x3x3]
%onnx::Conv_696[FLOAT, 24x72x1x1]
%onnx::Conv_699[FLOAT, 24x24x1x1]
%onnx::Conv_702[FLOAT, 24x1x5x5]
%onnx::Conv_705[FLOAT, 32x24x1x1]
%onnx::Conv_706[FLOAT, 32]
%onnx::Conv_708[FLOAT, 192x32x1x1]
%onnx::Conv_709[FLOAT, 192]
%onnx::Conv_711[FLOAT, 192x1x5x5]
%onnx::Conv_714[FLOAT, 32x192x1x1]
%onnx::Conv_717[FLOAT, 32x32x1x1]
%onnx::Conv_720[FLOAT, 32x1x3x3]
%onnx::Conv_723[FLOAT, 32x32x1x1]
%onnx::Conv_726[FLOAT, 32x32x1x1]
%onnx::Conv_729[FLOAT, 32x1x5x5]
%onnx::Conv_732[FLOAT, 32x32x1x1]
%onnx::Conv_735[FLOAT, 96x32x1x1]
%onnx::Conv_738[FLOAT, 96x1x3x3]
%onnx::Conv_741[FLOAT, 64x96x1x1]
%onnx::Conv_742[FLOAT, 64]
%onnx::Conv_744[FLOAT, 192x64x1x1]
%onnx::Conv_747[FLOAT, 192x1x3x3]
%onnx::Conv_750[FLOAT, 64x192x1x1]
%onnx::Conv_753[FLOAT, 64x32x1x1]
%onnx::Conv_756[FLOAT, 64x1x5x5]
%onnx::Conv_759[FLOAT, 64x32x1x1]
%onnx::Conv_762[FLOAT, 64x64x1x1]
%onnx::Conv_765[FLOAT, 64x1x5x5]
%onnx::Conv_768[FLOAT, 112x64x1x1]
%onnx::Conv_769[FLOAT, 112]
%onnx::Conv_771[FLOAT, 672x112x1x1]
%onnx::Conv_772[FLOAT, 672]
%onnx::Conv_774[FLOAT, 672x1x5x5]
%onnx::Conv_777[FLOAT, 112x672x1x1]
%onnx::Conv_780[FLOAT, 336x112x1x1]
%onnx::Conv_781[FLOAT, 336]
%onnx::Conv_783[FLOAT, 336x1x3x3]
%onnx::Conv_786[FLOAT, 112x336x1x1]
%onnx::Conv_789[FLOAT, 112x56x1x1]
%onnx::Conv_792[FLOAT, 112x1x3x3]
%onnx::Conv_795[FLOAT, 112x56x1x1]
%onnx::Conv_798[FLOAT, 184x112x1x1]
%onnx::Conv_799[FLOAT, 184]
%onnx::Conv_801[FLOAT, 552x184x1x1]
%onnx::Conv_802[FLOAT, 552]
%onnx::Conv_804[FLOAT, 552x1x5x5]
%onnx::Conv_807[FLOAT, 184x552x1x1]
%onnx::Conv_810[FLOAT, 184x184x1x1]
%onnx::Conv_813[FLOAT, 184x1x5x5]
%onnx::Conv_816[FLOAT, 184x184x1x1]
%onnx::Conv_819[FLOAT, 184x92x1x1]
%onnx::Conv_822[FLOAT, 184x1x5x5]
%onnx::Conv_825[FLOAT, 184x92x1x1]
%onnx::Conv_828[FLOAT, 184x184x1x1]
%onnx::Conv_831[FLOAT, 184x1x5x5]
%onnx::Conv_834[FLOAT, 352x184x1x1]
%onnx::Conv_835[FLOAT, 352]
%onnx::Conv_837[FLOAT, 1504x352x1x1]
%onnx::Conv_838[FLOAT, 1504]
) {
%onnx::Conv_832 = Identity(%onnx::Conv_799)
%onnx::Conv_829 = Identity(%onnx::Conv_799)
%onnx::Conv_826 = Identity(%onnx::Conv_799)
%onnx::Conv_823 = Identity(%onnx::Conv_799)
%onnx::Conv_820 = Identity(%onnx::Conv_799)
%onnx::Conv_817 = Identity(%onnx::Conv_799)
%onnx::Conv_814 = Identity(%onnx::Conv_799)
%onnx::Conv_811 = Identity(%onnx::Conv_799)
%onnx::Conv_808 = Identity(%onnx::Conv_799)
%onnx::Conv_805 = Identity(%onnx::Conv_802)
%onnx::Conv_796 = Identity(%onnx::Conv_769)
%onnx::Conv_793 = Identity(%onnx::Conv_769)
%onnx::Conv_790 = Identity(%onnx::Conv_769)
%onnx::Conv_787 = Identity(%onnx::Conv_769)
%onnx::Conv_784 = Identity(%onnx::Conv_781)
%onnx::Conv_778 = Identity(%onnx::Conv_769)
%onnx::Conv_775 = Identity(%onnx::Conv_772)
%onnx::Conv_766 = Identity(%onnx::Conv_742)
%onnx::Conv_763 = Identity(%onnx::Conv_742)
%onnx::Conv_760 = Identity(%onnx::Conv_742)
%onnx::Conv_757 = Identity(%onnx::Conv_742)
%onnx::Conv_754 = Identity(%onnx::Conv_742)
%onnx::Conv_751 = Identity(%onnx::Conv_742)
%onnx::Conv_748 = Identity(%onnx::Conv_709)
%onnx::Conv_745 = Identity(%onnx::Conv_709)
%onnx::Conv_739 = Identity(%onnx::Conv_655)
%onnx::Conv_736 = Identity(%onnx::Conv_655)
%onnx::Conv_733 = Identity(%onnx::Conv_706)
%onnx::Conv_730 = Identity(%onnx::Conv_706)
%onnx::Conv_727 = Identity(%onnx::Conv_706)
%onnx::Conv_724 = Identity(%onnx::Conv_706)
%onnx::Conv_721 = Identity(%onnx::Conv_706)
%onnx::Conv_718 = Identity(%onnx::Conv_706)
%onnx::Conv_715 = Identity(%onnx::Conv_706)
%onnx::Conv_712 = Identity(%onnx::Conv_709)
%onnx::Conv_703 = Identity(%onnx::Conv_670)
%onnx::Conv_700 = Identity(%onnx::Conv_670)
%onnx::Conv_697 = Identity(%onnx::Conv_670)
%onnx::Conv_694 = Identity(%onnx::Conv_691)
%onnx::Conv_688 = Identity(%onnx::Conv_670)
%onnx::Conv_685 = Identity(%onnx::Conv_670)
%onnx::Conv_682 = Identity(%onnx::Conv_670)
%onnx::Conv_679 = Identity(%onnx::Conv_670)
%onnx::Conv_676 = Identity(%onnx::Conv_670)
%onnx::Conv_673 = Identity(%onnx::Conv_670)
%onnx::Conv_667 = Identity(%onnx::Conv_652)
%onnx::Conv_664 = Identity(%onnx::Conv_652)
%onnx::Conv_661 = Identity(%onnx::Conv_652)
%onnx::Conv_658 = Identity(%onnx::Conv_655)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_651, %onnx::Conv_652)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%649 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %649
}
|
val_accuracy
| 0 | 58,285,312 | 1,501,412 |
{'zcp_synflow': 78.79970549683688, 'zcp_zen': 66.34237670898438, 'zcp_epe_nas': 18.692194999753593, 'zcp_fisher': 0.13068647682666779, 'zcp_flops': 58285312.0, 'zcp_grad_norm': 23.354042053222656, 'zcp_grasp': -0.041652679443359375, 'zcp_jacov': -16.06468220071104, 'zcp_l2_norm': 581.4013671875, 'zcp_nwot': 210.3767661738378, 'zcp_params': 1501412.0, 'zcp_plain': 0.0024586187209933996, 'zcp_snip': 39.424949645996094, 'lat_1080ti_1': 0.5750621667010484, 'lat_1080ti_32': 0.4684224798441301, 'lat_1080ti_64': 0.36839980559513935, 'lat_2080ti_1': 0.628242936474812, 'lat_2080ti_32': 0.49444001036797447, 'lat_2080ti_64': 0.41480406130986497, 'lat_essential_ph_1': 0.20754716981132076, 'lat_eyeriss': 0.3143417747052243, 'lat_fpga': 0.3523189423999842, 'lat_gold_6226': 0.2199639776476097, 'lat_gold_6240': 0.43500238232579563, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.3148529387894076, 'lat_raspi4': 0.3059286582943141, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.11811023622047244, 'lat_silver_4114': 0.4662805197360794, 'lat_silver_4210r': 0.49683184854821805, 'lat_titan_rtx_1': 0.5929512877283728, 'lat_titan_rtx_32': 0.4934619920513555, 'lat_titan_rtx_64': 0.44754720164093176, 'lat_titanx_1': 0.3143026763191469, 'lat_titanx_32': 0.47227957268105397, 'lat_titanx_64': 0.37005698033738577, 'lat_titanxp_1': 0.5549245706621221, 'lat_titanxp_32': 0.48220050134410886, 'lat_titanxp_64': 0.41600407471865314}
| |
FBNet_4920
|
FBNet
|
4920
|
4920
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_658[FLOAT, 16x3x3x3]
%onnx::Conv_659[FLOAT, 16]
%onnx::Conv_661[FLOAT, 16x8x1x1]
%onnx::Conv_664[FLOAT, 16x1x3x3]
%onnx::Conv_667[FLOAT, 16x8x1x1]
%onnx::Conv_670[FLOAT, 16x8x1x1]
%onnx::Conv_673[FLOAT, 16x1x5x5]
%onnx::Conv_676[FLOAT, 24x8x1x1]
%onnx::Conv_677[FLOAT, 24]
%onnx::Conv_679[FLOAT, 72x24x1x1]
%onnx::Conv_680[FLOAT, 72]
%onnx::Conv_682[FLOAT, 72x1x5x5]
%onnx::Conv_685[FLOAT, 24x72x1x1]
%onnx::Conv_688[FLOAT, 24x12x1x1]
%onnx::Conv_691[FLOAT, 24x1x5x5]
%onnx::Conv_694[FLOAT, 24x12x1x1]
%onnx::Conv_697[FLOAT, 24x24x1x1]
%onnx::Conv_700[FLOAT, 24x1x5x5]
%onnx::Conv_703[FLOAT, 24x24x1x1]
%onnx::Conv_706[FLOAT, 144x24x1x1]
%onnx::Conv_707[FLOAT, 144]
%onnx::Conv_709[FLOAT, 144x1x3x3]
%onnx::Conv_712[FLOAT, 32x144x1x1]
%onnx::Conv_713[FLOAT, 32]
%onnx::Conv_715[FLOAT, 192x32x1x1]
%onnx::Conv_716[FLOAT, 192]
%onnx::Conv_718[FLOAT, 192x1x5x5]
%onnx::Conv_721[FLOAT, 32x192x1x1]
%onnx::Conv_724[FLOAT, 32x32x1x1]
%onnx::Conv_727[FLOAT, 32x1x5x5]
%onnx::Conv_730[FLOAT, 32x32x1x1]
%onnx::Conv_733[FLOAT, 32x16x1x1]
%onnx::Conv_736[FLOAT, 32x1x5x5]
%onnx::Conv_739[FLOAT, 32x16x1x1]
%onnx::Conv_742[FLOAT, 32x16x1x1]
%onnx::Conv_745[FLOAT, 32x1x5x5]
%onnx::Conv_748[FLOAT, 64x16x1x1]
%onnx::Conv_749[FLOAT, 64]
%onnx::Conv_751[FLOAT, 384x64x1x1]
%onnx::Conv_752[FLOAT, 384]
%onnx::Conv_754[FLOAT, 384x1x5x5]
%onnx::Conv_757[FLOAT, 64x384x1x1]
%onnx::Conv_760[FLOAT, 192x64x1x1]
%onnx::Conv_763[FLOAT, 192x1x5x5]
%onnx::Conv_766[FLOAT, 64x192x1x1]
%onnx::Conv_769[FLOAT, 64x64x1x1]
%onnx::Conv_772[FLOAT, 64x1x3x3]
%onnx::Conv_775[FLOAT, 112x64x1x1]
%onnx::Conv_776[FLOAT, 112]
%onnx::Conv_778[FLOAT, 336x112x1x1]
%onnx::Conv_779[FLOAT, 336]
%onnx::Conv_781[FLOAT, 336x1x3x3]
%onnx::Conv_784[FLOAT, 112x336x1x1]
%onnx::Conv_787[FLOAT, 672x112x1x1]
%onnx::Conv_788[FLOAT, 672]
%onnx::Conv_790[FLOAT, 672x1x3x3]
%onnx::Conv_793[FLOAT, 112x672x1x1]
%onnx::Conv_796[FLOAT, 112x112x1x1]
%onnx::Conv_799[FLOAT, 112x1x3x3]
%onnx::Conv_802[FLOAT, 112x112x1x1]
%onnx::Conv_805[FLOAT, 672x112x1x1]
%onnx::Conv_808[FLOAT, 672x1x5x5]
%onnx::Conv_811[FLOAT, 184x672x1x1]
%onnx::Conv_812[FLOAT, 184]
%onnx::Conv_814[FLOAT, 184x184x1x1]
%onnx::Conv_817[FLOAT, 184x1x3x3]
%onnx::Conv_820[FLOAT, 184x184x1x1]
%onnx::Conv_823[FLOAT, 552x184x1x1]
%onnx::Conv_824[FLOAT, 552]
%onnx::Conv_826[FLOAT, 552x1x5x5]
%onnx::Conv_829[FLOAT, 184x552x1x1]
%onnx::Conv_832[FLOAT, 184x184x1x1]
%onnx::Conv_835[FLOAT, 184x1x5x5]
%onnx::Conv_838[FLOAT, 352x184x1x1]
%onnx::Conv_839[FLOAT, 352]
%onnx::Conv_841[FLOAT, 1504x352x1x1]
%onnx::Conv_842[FLOAT, 1504]
) {
%onnx::Conv_836 = Identity(%onnx::Conv_812)
%onnx::Conv_833 = Identity(%onnx::Conv_812)
%onnx::Conv_830 = Identity(%onnx::Conv_812)
%onnx::Conv_827 = Identity(%onnx::Conv_824)
%onnx::Conv_821 = Identity(%onnx::Conv_812)
%onnx::Conv_818 = Identity(%onnx::Conv_812)
%onnx::Conv_815 = Identity(%onnx::Conv_812)
%onnx::Conv_809 = Identity(%onnx::Conv_788)
%onnx::Conv_806 = Identity(%onnx::Conv_788)
%onnx::Conv_803 = Identity(%onnx::Conv_776)
%onnx::Conv_800 = Identity(%onnx::Conv_776)
%onnx::Conv_797 = Identity(%onnx::Conv_776)
%onnx::Conv_794 = Identity(%onnx::Conv_776)
%onnx::Conv_791 = Identity(%onnx::Conv_788)
%onnx::Conv_785 = Identity(%onnx::Conv_776)
%onnx::Conv_782 = Identity(%onnx::Conv_779)
%onnx::Conv_773 = Identity(%onnx::Conv_749)
%onnx::Conv_770 = Identity(%onnx::Conv_749)
%onnx::Conv_767 = Identity(%onnx::Conv_749)
%onnx::Conv_764 = Identity(%onnx::Conv_716)
%onnx::Conv_761 = Identity(%onnx::Conv_716)
%onnx::Conv_758 = Identity(%onnx::Conv_749)
%onnx::Conv_755 = Identity(%onnx::Conv_752)
%onnx::Conv_746 = Identity(%onnx::Conv_713)
%onnx::Conv_743 = Identity(%onnx::Conv_713)
%onnx::Conv_740 = Identity(%onnx::Conv_713)
%onnx::Conv_737 = Identity(%onnx::Conv_713)
%onnx::Conv_734 = Identity(%onnx::Conv_713)
%onnx::Conv_731 = Identity(%onnx::Conv_713)
%onnx::Conv_728 = Identity(%onnx::Conv_713)
%onnx::Conv_725 = Identity(%onnx::Conv_713)
%onnx::Conv_722 = Identity(%onnx::Conv_713)
%onnx::Conv_719 = Identity(%onnx::Conv_716)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_677)
%onnx::Conv_701 = Identity(%onnx::Conv_677)
%onnx::Conv_698 = Identity(%onnx::Conv_677)
%onnx::Conv_695 = Identity(%onnx::Conv_677)
%onnx::Conv_692 = Identity(%onnx::Conv_677)
%onnx::Conv_689 = Identity(%onnx::Conv_677)
%onnx::Conv_686 = Identity(%onnx::Conv_677)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_674 = Identity(%onnx::Conv_659)
%onnx::Conv_671 = Identity(%onnx::Conv_659)
%onnx::Conv_668 = Identity(%onnx::Conv_659)
%onnx::Conv_665 = Identity(%onnx::Conv_659)
%onnx::Conv_662 = Identity(%onnx::Conv_659)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_658, %onnx::Conv_659)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%656 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %656
}
|
val_accuracy
| 0 | 68,695,424 | 1,711,620 |
{'zcp_synflow': 76.8615255736478, 'zcp_zen': 66.63663482666016, 'zcp_epe_nas': 14.15105505836831, 'zcp_fisher': 0.0730937197804451, 'zcp_flops': 68695424.0, 'zcp_grad_norm': 24.55801010131836, 'zcp_grasp': -0.17355918884277344, 'zcp_jacov': -16.04876776692305, 'zcp_l2_norm': 601.70263671875, 'zcp_nwot': 211.27033988060904, 'zcp_params': 1711620.0, 'zcp_plain': -0.0044266777113080025, 'zcp_snip': 37.40144729614258, 'lat_1080ti_1': 0.5742422472881135, 'lat_1080ti_32': 0.5172039272475929, 'lat_1080ti_64': 0.4023922193167886, 'lat_2080ti_1': 0.6509104072697183, 'lat_2080ti_32': 0.5156147001348249, 'lat_2080ti_64': 0.42955990962172685, 'lat_essential_ph_1': 0.20754716981132076, 'lat_eyeriss': 0.40998111235398765, 'lat_fpga': 0.4110371169316932, 'lat_gold_6226': 0.3397212413312049, 'lat_gold_6240': 0.4773341412678341, 'lat_pixel2': 0.391304347826087, 'lat_pixel3': 0.44998657652356344, 'lat_raspi4': 0.4202102854608736, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.15748031496062992, 'lat_silver_4114': 0.501282037895008, 'lat_silver_4210r': 0.5418211543743223, 'lat_titan_rtx_1': 0.5880557812560386, 'lat_titan_rtx_32': 0.49140244745186196, 'lat_titan_rtx_64': 0.4352790292754746, 'lat_titanx_1': 0.30960292208157403, 'lat_titanx_32': 0.43583311070838815, 'lat_titanx_64': 0.3926818752917677, 'lat_titanxp_1': 0.5533474402890158, 'lat_titanxp_32': 0.4672469562800568, 'lat_titanxp_64': 0.42941019694363547}
| |
FBNet_591
|
FBNet
|
591
|
591
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_669[FLOAT, 16x3x3x3]
%onnx::Conv_670[FLOAT, 16]
%onnx::Conv_672[FLOAT, 16x16x1x1]
%onnx::Conv_675[FLOAT, 16x1x3x3]
%onnx::Conv_678[FLOAT, 16x16x1x1]
%onnx::Conv_681[FLOAT, 16x16x1x1]
%onnx::Conv_684[FLOAT, 16x1x5x5]
%onnx::Conv_687[FLOAT, 24x16x1x1]
%onnx::Conv_688[FLOAT, 24]
%onnx::Conv_690[FLOAT, 24x12x1x1]
%onnx::Conv_693[FLOAT, 24x1x3x3]
%onnx::Conv_696[FLOAT, 24x12x1x1]
%onnx::Conv_699[FLOAT, 24x24x1x1]
%onnx::Conv_702[FLOAT, 24x1x3x3]
%onnx::Conv_705[FLOAT, 24x24x1x1]
%onnx::Conv_708[FLOAT, 24x24x1x1]
%onnx::Conv_711[FLOAT, 24x1x3x3]
%onnx::Conv_714[FLOAT, 32x24x1x1]
%onnx::Conv_715[FLOAT, 32]
%onnx::Conv_717[FLOAT, 96x32x1x1]
%onnx::Conv_718[FLOAT, 96]
%onnx::Conv_720[FLOAT, 96x1x5x5]
%onnx::Conv_723[FLOAT, 32x96x1x1]
%onnx::Conv_726[FLOAT, 192x32x1x1]
%onnx::Conv_727[FLOAT, 192]
%onnx::Conv_729[FLOAT, 192x1x5x5]
%onnx::Conv_732[FLOAT, 32x192x1x1]
%onnx::Conv_735[FLOAT, 32x16x1x1]
%onnx::Conv_738[FLOAT, 32x1x3x3]
%onnx::Conv_741[FLOAT, 32x16x1x1]
%onnx::Conv_744[FLOAT, 96x32x1x1]
%onnx::Conv_747[FLOAT, 96x1x5x5]
%onnx::Conv_750[FLOAT, 64x96x1x1]
%onnx::Conv_751[FLOAT, 64]
%onnx::Conv_753[FLOAT, 192x64x1x1]
%onnx::Conv_756[FLOAT, 192x1x5x5]
%onnx::Conv_759[FLOAT, 64x192x1x1]
%onnx::Conv_762[FLOAT, 192x64x1x1]
%onnx::Conv_765[FLOAT, 192x1x3x3]
%onnx::Conv_768[FLOAT, 64x192x1x1]
%onnx::Conv_771[FLOAT, 64x32x1x1]
%onnx::Conv_774[FLOAT, 64x1x5x5]
%onnx::Conv_777[FLOAT, 112x32x1x1]
%onnx::Conv_778[FLOAT, 112]
%onnx::Conv_780[FLOAT, 112x112x1x1]
%onnx::Conv_783[FLOAT, 112x1x5x5]
%onnx::Conv_786[FLOAT, 112x112x1x1]
%onnx::Conv_789[FLOAT, 112x56x1x1]
%onnx::Conv_792[FLOAT, 112x1x3x3]
%onnx::Conv_795[FLOAT, 112x56x1x1]
%onnx::Conv_798[FLOAT, 112x56x1x1]
%onnx::Conv_801[FLOAT, 112x1x3x3]
%onnx::Conv_804[FLOAT, 184x56x1x1]
%onnx::Conv_805[FLOAT, 184]
%onnx::Conv_807[FLOAT, 184x92x1x1]
%onnx::Conv_810[FLOAT, 184x1x5x5]
%onnx::Conv_813[FLOAT, 184x92x1x1]
%onnx::Conv_816[FLOAT, 1104x184x1x1]
%onnx::Conv_817[FLOAT, 1104]
%onnx::Conv_819[FLOAT, 1104x1x3x3]
%onnx::Conv_822[FLOAT, 184x1104x1x1]
%onnx::Conv_825[FLOAT, 184x92x1x1]
%onnx::Conv_828[FLOAT, 184x1x5x5]
%onnx::Conv_831[FLOAT, 184x92x1x1]
%onnx::Conv_834[FLOAT, 552x184x1x1]
%onnx::Conv_835[FLOAT, 552]
%onnx::Conv_837[FLOAT, 552x1x5x5]
%onnx::Conv_840[FLOAT, 352x552x1x1]
%onnx::Conv_841[FLOAT, 352]
%onnx::Conv_843[FLOAT, 1504x352x1x1]
%onnx::Conv_844[FLOAT, 1504]
) {
%onnx::Conv_838 = Identity(%onnx::Conv_835)
%onnx::Conv_832 = Identity(%onnx::Conv_805)
%onnx::Conv_829 = Identity(%onnx::Conv_805)
%onnx::Conv_826 = Identity(%onnx::Conv_805)
%onnx::Conv_823 = Identity(%onnx::Conv_805)
%onnx::Conv_820 = Identity(%onnx::Conv_817)
%onnx::Conv_814 = Identity(%onnx::Conv_805)
%onnx::Conv_811 = Identity(%onnx::Conv_805)
%onnx::Conv_808 = Identity(%onnx::Conv_805)
%onnx::Conv_802 = Identity(%onnx::Conv_778)
%onnx::Conv_799 = Identity(%onnx::Conv_778)
%onnx::Conv_796 = Identity(%onnx::Conv_778)
%onnx::Conv_793 = Identity(%onnx::Conv_778)
%onnx::Conv_790 = Identity(%onnx::Conv_778)
%onnx::Conv_787 = Identity(%onnx::Conv_778)
%onnx::Conv_784 = Identity(%onnx::Conv_778)
%onnx::Conv_781 = Identity(%onnx::Conv_778)
%onnx::Conv_775 = Identity(%onnx::Conv_751)
%onnx::Conv_772 = Identity(%onnx::Conv_751)
%onnx::Conv_769 = Identity(%onnx::Conv_751)
%onnx::Conv_766 = Identity(%onnx::Conv_727)
%onnx::Conv_763 = Identity(%onnx::Conv_727)
%onnx::Conv_760 = Identity(%onnx::Conv_751)
%onnx::Conv_757 = Identity(%onnx::Conv_727)
%onnx::Conv_754 = Identity(%onnx::Conv_727)
%onnx::Conv_748 = Identity(%onnx::Conv_718)
%onnx::Conv_745 = Identity(%onnx::Conv_718)
%onnx::Conv_742 = Identity(%onnx::Conv_715)
%onnx::Conv_739 = Identity(%onnx::Conv_715)
%onnx::Conv_736 = Identity(%onnx::Conv_715)
%onnx::Conv_733 = Identity(%onnx::Conv_715)
%onnx::Conv_730 = Identity(%onnx::Conv_727)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%onnx::Conv_712 = Identity(%onnx::Conv_688)
%onnx::Conv_709 = Identity(%onnx::Conv_688)
%onnx::Conv_706 = Identity(%onnx::Conv_688)
%onnx::Conv_703 = Identity(%onnx::Conv_688)
%onnx::Conv_700 = Identity(%onnx::Conv_688)
%onnx::Conv_697 = Identity(%onnx::Conv_688)
%onnx::Conv_694 = Identity(%onnx::Conv_688)
%onnx::Conv_691 = Identity(%onnx::Conv_688)
%onnx::Conv_685 = Identity(%onnx::Conv_670)
%onnx::Conv_682 = Identity(%onnx::Conv_670)
%onnx::Conv_679 = Identity(%onnx::Conv_670)
%onnx::Conv_676 = Identity(%onnx::Conv_670)
%onnx::Conv_673 = Identity(%onnx::Conv_670)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_669, %onnx::Conv_670)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_output_0 = Reshape(%/cells.8/nl/Relu_output_0, %/cells.8/shuffle/Constant_output_0)
%/cells.8/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.8/shuffle/Reshape_output_0)
%/cells.8/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.8/shuffle/Reshape_1_output_0 = Reshape(%/cells.8/shuffle/Transpose_output_0, %/cells.8/shuffle/Constant_1_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%667 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %667
}
|
val_accuracy
| 0 | 44,034,688 | 1,669,060 |
{'zcp_synflow': 69.12914884316403, 'zcp_zen': 60.05060958862305, 'zcp_epe_nas': 6.36738739641051, 'zcp_fisher': 0.06355752050876617, 'zcp_flops': 44034688.0, 'zcp_grad_norm': 17.642459869384766, 'zcp_grasp': 0.030877113342285156, 'zcp_jacov': -16.056767228450177, 'zcp_l2_norm': 520.5380859375, 'zcp_nwot': 203.072728895767, 'zcp_params': 1669060.0, 'zcp_plain': 0.004708330146968365, 'zcp_snip': 32.17304229736328, 'lat_1080ti_1': 0.5100930263519633, 'lat_1080ti_32': 0.4123916920637211, 'lat_1080ti_64': 0.17340696033875208, 'lat_2080ti_1': 0.5368862439356069, 'lat_2080ti_32': 0.3976895708759022, 'lat_2080ti_64': 0.21431199454669647, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.17079566501017723, 'lat_fpga': 0.1701679217091006, 'lat_gold_6226': 0.20729031430014042, 'lat_gold_6240': 0.3672129585118869, 'lat_pixel2': 0.13043478260869565, 'lat_pixel3': 0.16493963493816755, 'lat_raspi4': 0.27319964524774915, 'lat_samsung_a50': 0.08421052631578947, 'lat_samsung_s7': 0.09448818897637795, 'lat_silver_4114': 0.3888392650553905, 'lat_silver_4210r': 0.3918748328879434, 'lat_titan_rtx_1': 0.5167619370414945, 'lat_titan_rtx_32': 0.4165923224970784, 'lat_titan_rtx_64': 0.2633543476796788, 'lat_titanx_1': 0.2760123375127219, 'lat_titanx_32': 0.30829172175776903, 'lat_titanx_64': 0.1737179069440626, 'lat_titanxp_1': 0.48060969832522055, 'lat_titanxp_32': 0.3709561202606596, 'lat_titanxp_64': 0.20783682385801466}
| |
FBNet_4046
|
FBNet
|
4046
|
4046
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_615[FLOAT, 16x3x3x3]
%onnx::Conv_616[FLOAT, 16]
%onnx::Conv_618[FLOAT, 96x16x1x1]
%onnx::Conv_619[FLOAT, 96]
%onnx::Conv_621[FLOAT, 96x1x3x3]
%onnx::Conv_624[FLOAT, 16x96x1x1]
%onnx::Conv_627[FLOAT, 16x8x1x1]
%onnx::Conv_630[FLOAT, 16x1x3x3]
%onnx::Conv_633[FLOAT, 24x8x1x1]
%onnx::Conv_634[FLOAT, 24]
%onnx::Conv_636[FLOAT, 72x24x1x1]
%onnx::Conv_637[FLOAT, 72]
%onnx::Conv_639[FLOAT, 72x1x3x3]
%onnx::Conv_642[FLOAT, 24x72x1x1]
%onnx::Conv_645[FLOAT, 24x12x1x1]
%onnx::Conv_648[FLOAT, 24x1x5x5]
%onnx::Conv_651[FLOAT, 24x12x1x1]
%onnx::Conv_654[FLOAT, 24x12x1x1]
%onnx::Conv_657[FLOAT, 24x1x5x5]
%onnx::Conv_660[FLOAT, 24x12x1x1]
%onnx::Conv_663[FLOAT, 24x12x1x1]
%onnx::Conv_666[FLOAT, 24x1x5x5]
%onnx::Conv_669[FLOAT, 32x12x1x1]
%onnx::Conv_670[FLOAT, 32]
%onnx::Conv_672[FLOAT, 32x16x1x1]
%onnx::Conv_675[FLOAT, 32x1x5x5]
%onnx::Conv_678[FLOAT, 32x16x1x1]
%onnx::Conv_681[FLOAT, 192x32x1x1]
%onnx::Conv_682[FLOAT, 192]
%onnx::Conv_684[FLOAT, 192x1x3x3]
%onnx::Conv_687[FLOAT, 32x192x1x1]
%onnx::Conv_690[FLOAT, 96x32x1x1]
%onnx::Conv_693[FLOAT, 96x1x3x3]
%onnx::Conv_696[FLOAT, 64x96x1x1]
%onnx::Conv_697[FLOAT, 64]
%onnx::Conv_699[FLOAT, 64x64x1x1]
%onnx::Conv_702[FLOAT, 64x1x5x5]
%onnx::Conv_705[FLOAT, 64x64x1x1]
%onnx::Conv_708[FLOAT, 192x64x1x1]
%onnx::Conv_711[FLOAT, 192x1x5x5]
%onnx::Conv_714[FLOAT, 64x192x1x1]
%onnx::Conv_717[FLOAT, 384x64x1x1]
%onnx::Conv_718[FLOAT, 384]
%onnx::Conv_720[FLOAT, 384x1x5x5]
%onnx::Conv_723[FLOAT, 64x384x1x1]
%onnx::Conv_726[FLOAT, 192x64x1x1]
%onnx::Conv_729[FLOAT, 192x1x3x3]
%onnx::Conv_732[FLOAT, 112x192x1x1]
%onnx::Conv_733[FLOAT, 112]
%onnx::Conv_735[FLOAT, 112x56x1x1]
%onnx::Conv_738[FLOAT, 112x1x3x3]
%onnx::Conv_741[FLOAT, 112x56x1x1]
%onnx::Conv_744[FLOAT, 112x112x1x1]
%onnx::Conv_747[FLOAT, 112x1x3x3]
%onnx::Conv_750[FLOAT, 184x112x1x1]
%onnx::Conv_751[FLOAT, 184]
%onnx::Conv_753[FLOAT, 552x184x1x1]
%onnx::Conv_754[FLOAT, 552]
%onnx::Conv_756[FLOAT, 552x1x3x3]
%onnx::Conv_759[FLOAT, 184x552x1x1]
%onnx::Conv_762[FLOAT, 184x92x1x1]
%onnx::Conv_765[FLOAT, 184x1x3x3]
%onnx::Conv_768[FLOAT, 352x92x1x1]
%onnx::Conv_769[FLOAT, 352]
%onnx::Conv_771[FLOAT, 1504x352x1x1]
%onnx::Conv_772[FLOAT, 1504]
) {
%onnx::Conv_766 = Identity(%onnx::Conv_751)
%onnx::Conv_763 = Identity(%onnx::Conv_751)
%onnx::Conv_760 = Identity(%onnx::Conv_751)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_748 = Identity(%onnx::Conv_733)
%onnx::Conv_745 = Identity(%onnx::Conv_733)
%onnx::Conv_742 = Identity(%onnx::Conv_733)
%onnx::Conv_739 = Identity(%onnx::Conv_733)
%onnx::Conv_736 = Identity(%onnx::Conv_733)
%onnx::Conv_730 = Identity(%onnx::Conv_682)
%onnx::Conv_727 = Identity(%onnx::Conv_682)
%onnx::Conv_724 = Identity(%onnx::Conv_697)
%onnx::Conv_721 = Identity(%onnx::Conv_718)
%onnx::Conv_715 = Identity(%onnx::Conv_697)
%onnx::Conv_712 = Identity(%onnx::Conv_682)
%onnx::Conv_709 = Identity(%onnx::Conv_682)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_697)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%onnx::Conv_694 = Identity(%onnx::Conv_619)
%onnx::Conv_691 = Identity(%onnx::Conv_619)
%onnx::Conv_688 = Identity(%onnx::Conv_670)
%onnx::Conv_685 = Identity(%onnx::Conv_682)
%onnx::Conv_679 = Identity(%onnx::Conv_670)
%onnx::Conv_676 = Identity(%onnx::Conv_670)
%onnx::Conv_673 = Identity(%onnx::Conv_670)
%onnx::Conv_667 = Identity(%onnx::Conv_634)
%onnx::Conv_664 = Identity(%onnx::Conv_634)
%onnx::Conv_661 = Identity(%onnx::Conv_634)
%onnx::Conv_658 = Identity(%onnx::Conv_634)
%onnx::Conv_655 = Identity(%onnx::Conv_634)
%onnx::Conv_652 = Identity(%onnx::Conv_634)
%onnx::Conv_649 = Identity(%onnx::Conv_634)
%onnx::Conv_646 = Identity(%onnx::Conv_634)
%onnx::Conv_643 = Identity(%onnx::Conv_634)
%onnx::Conv_640 = Identity(%onnx::Conv_637)
%onnx::Conv_631 = Identity(%onnx::Conv_616)
%onnx::Conv_628 = Identity(%onnx::Conv_616)
%onnx::Conv_625 = Identity(%onnx::Conv_616)
%onnx::Conv_622 = Identity(%onnx::Conv_619)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_615, %onnx::Conv_616)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_output_0 = Reshape(%/cells.3/nl/Relu_output_0, %/cells.3/shuffle/Constant_output_0)
%/cells.3/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.3/shuffle/Reshape_output_0)
%/cells.3/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.3/shuffle/Reshape_1_output_0 = Reshape(%/cells.3/shuffle/Transpose_output_0, %/cells.3/shuffle/Constant_1_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/shuffle/Reshape_1_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_output_0 = Reshape(%/cells.6/nl/Relu_output_0, %/cells.6/shuffle/Constant_output_0)
%/cells.6/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.6/shuffle/Reshape_output_0)
%/cells.6/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.6/shuffle/Reshape_1_output_0 = Reshape(%/cells.6/shuffle/Transpose_output_0, %/cells.6/shuffle/Constant_1_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/shuffle/Reshape_1_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%613 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %613
}
|
val_accuracy
| 0 | 43,216,768 | 1,173,940 |
{'zcp_synflow': 60.2942691839112, 'zcp_zen': 52.38585662841797, 'zcp_epe_nas': 16.320425320738885, 'zcp_fisher': 0.07050113379955292, 'zcp_flops': 43216768.0, 'zcp_grad_norm': 16.517057418823242, 'zcp_grasp': -0.060973167419433594, 'zcp_jacov': -16.05426273373708, 'zcp_l2_norm': 443.16180419921875, 'zcp_nwot': 208.7012927015012, 'zcp_params': 1173940.0, 'zcp_plain': 0.0037879743613302708, 'zcp_snip': 27.741897583007812, 'lat_1080ti_1': 0.37720776055993666, 'lat_1080ti_32': 0.2566394180191689, 'lat_1080ti_64': 0.23408557460115564, 'lat_2080ti_1': 0.36486651591953, 'lat_2080ti_32': 0.3242344265026832, 'lat_2080ti_64': 0.2697001976477127, 'lat_essential_ph_1': 0.1320754716981132, 'lat_eyeriss': 0.17931344323620563, 'lat_fpga': 0.1520201070661692, 'lat_gold_6226': 0.14689984578748672, 'lat_gold_6240': 0.2636792110988819, 'lat_pixel2': 0.08695652173913043, 'lat_pixel3': 0.21067686767588575, 'lat_raspi4': 0.202761073082, 'lat_samsung_a50': 0.07368421052631578, 'lat_samsung_s7': 0.023622047244094488, 'lat_silver_4114': 0.2598875887786878, 'lat_silver_4210r': 0.2993952341771371, 'lat_titan_rtx_1': 0.32810962617408856, 'lat_titan_rtx_32': 0.2911803306253366, 'lat_titan_rtx_64': 0.2467206270868986, 'lat_titanx_1': 0.1697336548383702, 'lat_titanx_32': 0.2306424531259287, 'lat_titanx_64': 0.28281477780289027, 'lat_titanxp_1': 0.31747386831703533, 'lat_titanxp_32': 0.2580022497091372, 'lat_titanxp_64': 0.2559812487365829}
| |
FBNet_4831
|
FBNet
|
4831
|
4831
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_631[FLOAT, 16x3x3x3]
%onnx::Conv_632[FLOAT, 16]
%onnx::Conv_634[FLOAT, 48x16x1x1]
%onnx::Conv_635[FLOAT, 48]
%onnx::Conv_637[FLOAT, 48x1x5x5]
%onnx::Conv_640[FLOAT, 16x48x1x1]
%onnx::Conv_643[FLOAT, 96x16x1x1]
%onnx::Conv_644[FLOAT, 96]
%onnx::Conv_646[FLOAT, 96x1x3x3]
%onnx::Conv_649[FLOAT, 24x96x1x1]
%onnx::Conv_650[FLOAT, 24]
%onnx::Conv_652[FLOAT, 72x24x1x1]
%onnx::Conv_653[FLOAT, 72]
%onnx::Conv_655[FLOAT, 72x1x3x3]
%onnx::Conv_658[FLOAT, 24x72x1x1]
%onnx::Conv_661[FLOAT, 144x24x1x1]
%onnx::Conv_662[FLOAT, 144]
%onnx::Conv_664[FLOAT, 144x1x5x5]
%onnx::Conv_667[FLOAT, 24x144x1x1]
%onnx::Conv_670[FLOAT, 144x24x1x1]
%onnx::Conv_673[FLOAT, 144x1x5x5]
%onnx::Conv_676[FLOAT, 24x144x1x1]
%onnx::Conv_679[FLOAT, 72x24x1x1]
%onnx::Conv_682[FLOAT, 72x1x3x3]
%onnx::Conv_685[FLOAT, 32x72x1x1]
%onnx::Conv_686[FLOAT, 32]
%onnx::Conv_688[FLOAT, 32x32x1x1]
%onnx::Conv_691[FLOAT, 32x1x3x3]
%onnx::Conv_694[FLOAT, 32x32x1x1]
%onnx::Conv_697[FLOAT, 32x32x1x1]
%onnx::Conv_700[FLOAT, 32x1x3x3]
%onnx::Conv_703[FLOAT, 32x32x1x1]
%onnx::Conv_706[FLOAT, 32x32x1x1]
%onnx::Conv_709[FLOAT, 32x1x5x5]
%onnx::Conv_712[FLOAT, 32x32x1x1]
%onnx::Conv_715[FLOAT, 32x16x1x1]
%onnx::Conv_718[FLOAT, 32x1x3x3]
%onnx::Conv_721[FLOAT, 64x16x1x1]
%onnx::Conv_722[FLOAT, 64]
%onnx::Conv_724[FLOAT, 64x32x1x1]
%onnx::Conv_727[FLOAT, 64x1x5x5]
%onnx::Conv_730[FLOAT, 64x32x1x1]
%onnx::Conv_733[FLOAT, 384x64x1x1]
%onnx::Conv_734[FLOAT, 384]
%onnx::Conv_736[FLOAT, 384x1x5x5]
%onnx::Conv_739[FLOAT, 64x384x1x1]
%onnx::Conv_742[FLOAT, 384x64x1x1]
%onnx::Conv_745[FLOAT, 384x1x3x3]
%onnx::Conv_748[FLOAT, 64x384x1x1]
%onnx::Conv_751[FLOAT, 64x32x1x1]
%onnx::Conv_754[FLOAT, 64x1x3x3]
%onnx::Conv_757[FLOAT, 112x32x1x1]
%onnx::Conv_758[FLOAT, 112]
%onnx::Conv_760[FLOAT, 112x112x1x1]
%onnx::Conv_763[FLOAT, 112x1x3x3]
%onnx::Conv_766[FLOAT, 112x112x1x1]
%onnx::Conv_769[FLOAT, 672x112x1x1]
%onnx::Conv_770[FLOAT, 672]
%onnx::Conv_772[FLOAT, 672x1x3x3]
%onnx::Conv_775[FLOAT, 112x672x1x1]
%onnx::Conv_778[FLOAT, 672x112x1x1]
%onnx::Conv_781[FLOAT, 672x1x5x5]
%onnx::Conv_784[FLOAT, 184x672x1x1]
%onnx::Conv_785[FLOAT, 184]
%onnx::Conv_787[FLOAT, 552x184x1x1]
%onnx::Conv_788[FLOAT, 552]
%onnx::Conv_790[FLOAT, 552x1x3x3]
%onnx::Conv_793[FLOAT, 184x552x1x1]
%onnx::Conv_796[FLOAT, 552x184x1x1]
%onnx::Conv_799[FLOAT, 552x1x5x5]
%onnx::Conv_802[FLOAT, 184x552x1x1]
%onnx::Conv_805[FLOAT, 184x184x1x1]
%onnx::Conv_808[FLOAT, 184x1x3x3]
%onnx::Conv_811[FLOAT, 184x184x1x1]
%onnx::Conv_814[FLOAT, 352x184x1x1]
%onnx::Conv_815[FLOAT, 352]
%onnx::Conv_817[FLOAT, 1504x352x1x1]
%onnx::Conv_818[FLOAT, 1504]
) {
%onnx::Conv_812 = Identity(%onnx::Conv_785)
%onnx::Conv_809 = Identity(%onnx::Conv_785)
%onnx::Conv_806 = Identity(%onnx::Conv_785)
%onnx::Conv_803 = Identity(%onnx::Conv_785)
%onnx::Conv_800 = Identity(%onnx::Conv_788)
%onnx::Conv_797 = Identity(%onnx::Conv_788)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_788)
%onnx::Conv_782 = Identity(%onnx::Conv_770)
%onnx::Conv_779 = Identity(%onnx::Conv_770)
%onnx::Conv_776 = Identity(%onnx::Conv_758)
%onnx::Conv_773 = Identity(%onnx::Conv_770)
%onnx::Conv_767 = Identity(%onnx::Conv_758)
%onnx::Conv_764 = Identity(%onnx::Conv_758)
%onnx::Conv_761 = Identity(%onnx::Conv_758)
%onnx::Conv_755 = Identity(%onnx::Conv_722)
%onnx::Conv_752 = Identity(%onnx::Conv_722)
%onnx::Conv_749 = Identity(%onnx::Conv_722)
%onnx::Conv_746 = Identity(%onnx::Conv_734)
%onnx::Conv_743 = Identity(%onnx::Conv_734)
%onnx::Conv_740 = Identity(%onnx::Conv_722)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_731 = Identity(%onnx::Conv_722)
%onnx::Conv_728 = Identity(%onnx::Conv_722)
%onnx::Conv_725 = Identity(%onnx::Conv_722)
%onnx::Conv_719 = Identity(%onnx::Conv_686)
%onnx::Conv_716 = Identity(%onnx::Conv_686)
%onnx::Conv_713 = Identity(%onnx::Conv_686)
%onnx::Conv_710 = Identity(%onnx::Conv_686)
%onnx::Conv_707 = Identity(%onnx::Conv_686)
%onnx::Conv_704 = Identity(%onnx::Conv_686)
%onnx::Conv_701 = Identity(%onnx::Conv_686)
%onnx::Conv_698 = Identity(%onnx::Conv_686)
%onnx::Conv_695 = Identity(%onnx::Conv_686)
%onnx::Conv_692 = Identity(%onnx::Conv_686)
%onnx::Conv_689 = Identity(%onnx::Conv_686)
%onnx::Conv_683 = Identity(%onnx::Conv_653)
%onnx::Conv_680 = Identity(%onnx::Conv_653)
%onnx::Conv_677 = Identity(%onnx::Conv_650)
%onnx::Conv_674 = Identity(%onnx::Conv_662)
%onnx::Conv_671 = Identity(%onnx::Conv_662)
%onnx::Conv_668 = Identity(%onnx::Conv_650)
%onnx::Conv_665 = Identity(%onnx::Conv_662)
%onnx::Conv_659 = Identity(%onnx::Conv_650)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_647 = Identity(%onnx::Conv_644)
%onnx::Conv_641 = Identity(%onnx::Conv_632)
%onnx::Conv_638 = Identity(%onnx::Conv_635)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_631, %onnx::Conv_632)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_634, %onnx::Conv_635)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%629 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %629
}
|
val_accuracy
| 0 | 86,913,152 | 1,832,012 |
{'zcp_synflow': 79.47276352067566, 'zcp_zen': 68.88404846191406, 'zcp_epe_nas': 6.4892919606390445, 'zcp_fisher': 0.12531417608261108, 'zcp_flops': 86913152.0, 'zcp_grad_norm': 24.266096115112305, 'zcp_grasp': -0.1215972900390625, 'zcp_jacov': -16.05403384544669, 'zcp_l2_norm': 644.1441040039062, 'zcp_nwot': 218.62366247858358, 'zcp_params': 1832012.0, 'zcp_plain': 0.002439548261463642, 'zcp_snip': 45.9183235168457, 'lat_1080ti_1': 0.5715917572682305, 'lat_1080ti_32': 0.7468021693329515, 'lat_1080ti_64': 0.7656544952663115, 'lat_2080ti_1': 0.6391650641830396, 'lat_2080ti_32': 0.7485040258181923, 'lat_2080ti_64': 0.7849144077943208, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.6823024590614856, 'lat_fpga': 0.6247142758190761, 'lat_gold_6226': 0.40460040514287055, 'lat_gold_6240': 0.4622957994818573, 'lat_pixel2': 0.5217391304347826, 'lat_pixel3': 0.6933914718436514, 'lat_raspi4': 0.6499537417582866, 'lat_samsung_a50': 0.2631578947368421, 'lat_samsung_s7': 0.2047244094488189, 'lat_silver_4114': 0.47860180333817665, 'lat_silver_4210r': 0.49694063475087247, 'lat_titan_rtx_1': 0.593466883329192, 'lat_titan_rtx_32': 0.6577473852236784, 'lat_titan_rtx_64': 0.7695894661406285, 'lat_titanx_1': 0.3158228176439548, 'lat_titanx_32': 0.8168567187514715, 'lat_titanx_64': 0.7710992322880439, 'lat_titanxp_1': 0.5576184926927586, 'lat_titanxp_32': 0.7620755057644841, 'lat_titanxp_64': 0.7982363448317489}
| |
FBNet_976
|
FBNet
|
976
|
976
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_597[FLOAT, 16x3x3x3]
%onnx::Conv_598[FLOAT, 16]
%onnx::Conv_600[FLOAT, 16x8x1x1]
%onnx::Conv_603[FLOAT, 16x1x3x3]
%onnx::Conv_606[FLOAT, 16x8x1x1]
%onnx::Conv_609[FLOAT, 96x16x1x1]
%onnx::Conv_610[FLOAT, 96]
%onnx::Conv_612[FLOAT, 96x1x3x3]
%onnx::Conv_615[FLOAT, 24x96x1x1]
%onnx::Conv_616[FLOAT, 24]
%onnx::Conv_618[FLOAT, 144x24x1x1]
%onnx::Conv_619[FLOAT, 144]
%onnx::Conv_621[FLOAT, 144x1x5x5]
%onnx::Conv_624[FLOAT, 24x144x1x1]
%onnx::Conv_627[FLOAT, 24x12x1x1]
%onnx::Conv_630[FLOAT, 24x1x3x3]
%onnx::Conv_633[FLOAT, 24x12x1x1]
%onnx::Conv_636[FLOAT, 24x24x1x1]
%onnx::Conv_639[FLOAT, 24x1x3x3]
%onnx::Conv_642[FLOAT, 32x24x1x1]
%onnx::Conv_643[FLOAT, 32]
%onnx::Conv_645[FLOAT, 96x32x1x1]
%onnx::Conv_648[FLOAT, 96x1x3x3]
%onnx::Conv_651[FLOAT, 32x96x1x1]
%onnx::Conv_654[FLOAT, 32x32x1x1]
%onnx::Conv_657[FLOAT, 32x1x5x5]
%onnx::Conv_660[FLOAT, 32x32x1x1]
%onnx::Conv_663[FLOAT, 32x16x1x1]
%onnx::Conv_666[FLOAT, 32x1x3x3]
%onnx::Conv_669[FLOAT, 64x16x1x1]
%onnx::Conv_670[FLOAT, 64]
%onnx::Conv_672[FLOAT, 384x64x1x1]
%onnx::Conv_673[FLOAT, 384]
%onnx::Conv_675[FLOAT, 384x1x3x3]
%onnx::Conv_678[FLOAT, 64x384x1x1]
%onnx::Conv_681[FLOAT, 384x64x1x1]
%onnx::Conv_684[FLOAT, 384x1x3x3]
%onnx::Conv_687[FLOAT, 64x384x1x1]
%onnx::Conv_690[FLOAT, 384x64x1x1]
%onnx::Conv_693[FLOAT, 384x1x5x5]
%onnx::Conv_696[FLOAT, 112x384x1x1]
%onnx::Conv_697[FLOAT, 112]
%onnx::Conv_699[FLOAT, 112x56x1x1]
%onnx::Conv_702[FLOAT, 112x1x3x3]
%onnx::Conv_705[FLOAT, 112x56x1x1]
%onnx::Conv_708[FLOAT, 112x56x1x1]
%onnx::Conv_711[FLOAT, 112x1x5x5]
%onnx::Conv_714[FLOAT, 184x56x1x1]
%onnx::Conv_715[FLOAT, 184]
%onnx::Conv_717[FLOAT, 184x184x1x1]
%onnx::Conv_720[FLOAT, 184x1x5x5]
%onnx::Conv_723[FLOAT, 184x184x1x1]
%onnx::Conv_726[FLOAT, 1104x184x1x1]
%onnx::Conv_727[FLOAT, 1104]
%onnx::Conv_729[FLOAT, 1104x1x5x5]
%onnx::Conv_732[FLOAT, 184x1104x1x1]
%onnx::Conv_735[FLOAT, 184x92x1x1]
%onnx::Conv_738[FLOAT, 184x1x3x3]
%onnx::Conv_741[FLOAT, 184x92x1x1]
%onnx::Conv_744[FLOAT, 184x184x1x1]
%onnx::Conv_747[FLOAT, 184x1x3x3]
%onnx::Conv_750[FLOAT, 352x184x1x1]
%onnx::Conv_751[FLOAT, 352]
%onnx::Conv_753[FLOAT, 1504x352x1x1]
%onnx::Conv_754[FLOAT, 1504]
) {
%onnx::Conv_748 = Identity(%onnx::Conv_715)
%onnx::Conv_745 = Identity(%onnx::Conv_715)
%onnx::Conv_742 = Identity(%onnx::Conv_715)
%onnx::Conv_739 = Identity(%onnx::Conv_715)
%onnx::Conv_736 = Identity(%onnx::Conv_715)
%onnx::Conv_733 = Identity(%onnx::Conv_715)
%onnx::Conv_730 = Identity(%onnx::Conv_727)
%onnx::Conv_724 = Identity(%onnx::Conv_715)
%onnx::Conv_721 = Identity(%onnx::Conv_715)
%onnx::Conv_718 = Identity(%onnx::Conv_715)
%onnx::Conv_712 = Identity(%onnx::Conv_697)
%onnx::Conv_709 = Identity(%onnx::Conv_697)
%onnx::Conv_706 = Identity(%onnx::Conv_697)
%onnx::Conv_703 = Identity(%onnx::Conv_697)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%onnx::Conv_694 = Identity(%onnx::Conv_673)
%onnx::Conv_691 = Identity(%onnx::Conv_673)
%onnx::Conv_688 = Identity(%onnx::Conv_670)
%onnx::Conv_685 = Identity(%onnx::Conv_673)
%onnx::Conv_682 = Identity(%onnx::Conv_673)
%onnx::Conv_679 = Identity(%onnx::Conv_670)
%onnx::Conv_676 = Identity(%onnx::Conv_673)
%onnx::Conv_667 = Identity(%onnx::Conv_643)
%onnx::Conv_664 = Identity(%onnx::Conv_643)
%onnx::Conv_661 = Identity(%onnx::Conv_643)
%onnx::Conv_658 = Identity(%onnx::Conv_643)
%onnx::Conv_655 = Identity(%onnx::Conv_643)
%onnx::Conv_652 = Identity(%onnx::Conv_643)
%onnx::Conv_649 = Identity(%onnx::Conv_610)
%onnx::Conv_646 = Identity(%onnx::Conv_610)
%onnx::Conv_640 = Identity(%onnx::Conv_616)
%onnx::Conv_637 = Identity(%onnx::Conv_616)
%onnx::Conv_634 = Identity(%onnx::Conv_616)
%onnx::Conv_631 = Identity(%onnx::Conv_616)
%onnx::Conv_628 = Identity(%onnx::Conv_616)
%onnx::Conv_625 = Identity(%onnx::Conv_616)
%onnx::Conv_622 = Identity(%onnx::Conv_619)
%onnx::Conv_613 = Identity(%onnx::Conv_610)
%onnx::Conv_607 = Identity(%onnx::Conv_598)
%onnx::Conv_604 = Identity(%onnx::Conv_598)
%onnx::Conv_601 = Identity(%onnx::Conv_598)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_597, %onnx::Conv_598)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_600, %onnx::Conv_601)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_603, %onnx::Conv_604)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_606, %onnx::Conv_607)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_609, %onnx::Conv_610)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_612, %onnx::Conv_613)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_615, %onnx::Conv_616)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_output_0 = Reshape(%/cells.17/nl/Relu_output_0, %/cells.17/shuffle/Constant_output_0)
%/cells.17/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.17/shuffle/Reshape_output_0)
%/cells.17/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.17/shuffle/Reshape_1_output_0 = Reshape(%/cells.17/shuffle/Transpose_output_0, %/cells.17/shuffle/Constant_1_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/shuffle/Reshape_1_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%595 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %595
}
|
val_accuracy
| 0 | 55,575,680 | 1,587,692 |
{'zcp_synflow': 62.00459912406989, 'zcp_zen': 55.60196304321289, 'zcp_epe_nas': 27.5485744025104, 'zcp_fisher': 0.0576866939663887, 'zcp_flops': 55575680.0, 'zcp_grad_norm': 16.998563766479492, 'zcp_grasp': -0.07044410705566406, 'zcp_jacov': -16.04504680295001, 'zcp_l2_norm': 506.84027099609375, 'zcp_nwot': 210.8193093344619, 'zcp_params': 1587692.0, 'zcp_plain': -0.0029724985361099243, 'zcp_snip': 29.2490177154541, 'lat_1080ti_1': 0.3458024298210271, 'lat_1080ti_32': 0.40944594725966477, 'lat_1080ti_64': 0.33547954693969606, 'lat_2080ti_1': 0.3287336260586523, 'lat_2080ti_32': 0.3834543273247919, 'lat_2080ti_64': 0.3364503817169364, 'lat_essential_ph_1': 0.32075471698113206, 'lat_eyeriss': 0.33675022463462484, 'lat_fpga': 0.30980912140432804, 'lat_gold_6226': 0.23924826192464102, 'lat_gold_6240': 0.32102192566103865, 'lat_pixel2': 0.2391304347826087, 'lat_pixel3': 0.337116092844471, 'lat_raspi4': 0.34794541174605986, 'lat_samsung_a50': 0.21052631578947367, 'lat_samsung_s7': 0.15748031496062992, 'lat_silver_4114': 0.3086665706596688, 'lat_silver_4210r': 0.29192772449109555, 'lat_titan_rtx_1': 0.310778382990149, 'lat_titan_rtx_32': 0.3693111285642442, 'lat_titan_rtx_64': 0.35033437628414305, 'lat_titanx_1': 0.15024407655221333, 'lat_titanx_32': 0.35003257429840956, 'lat_titanx_64': 0.38437088220146143, 'lat_titanxp_1': 0.3004755141965758, 'lat_titanxp_32': 0.360513386025983, 'lat_titanxp_64': 0.3547300889502218}
| |
FBNet_4424
|
FBNet
|
4424
|
4424
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_715[FLOAT, 16x3x3x3]
%onnx::Conv_716[FLOAT, 16]
%onnx::Conv_718[FLOAT, 48x16x1x1]
%onnx::Conv_719[FLOAT, 48]
%onnx::Conv_721[FLOAT, 48x1x3x3]
%onnx::Conv_724[FLOAT, 16x48x1x1]
%onnx::Conv_727[FLOAT, 16x8x1x1]
%onnx::Conv_730[FLOAT, 16x1x3x3]
%onnx::Conv_733[FLOAT, 24x8x1x1]
%onnx::Conv_734[FLOAT, 24]
%onnx::Conv_736[FLOAT, 72x24x1x1]
%onnx::Conv_737[FLOAT, 72]
%onnx::Conv_739[FLOAT, 72x1x3x3]
%onnx::Conv_742[FLOAT, 24x72x1x1]
%onnx::Conv_745[FLOAT, 144x24x1x1]
%onnx::Conv_746[FLOAT, 144]
%onnx::Conv_748[FLOAT, 144x1x5x5]
%onnx::Conv_751[FLOAT, 24x144x1x1]
%onnx::Conv_754[FLOAT, 24x12x1x1]
%onnx::Conv_757[FLOAT, 24x1x5x5]
%onnx::Conv_760[FLOAT, 32x12x1x1]
%onnx::Conv_761[FLOAT, 32]
%onnx::Conv_763[FLOAT, 192x32x1x1]
%onnx::Conv_764[FLOAT, 192]
%onnx::Conv_766[FLOAT, 192x1x5x5]
%onnx::Conv_769[FLOAT, 32x192x1x1]
%onnx::Conv_772[FLOAT, 32x16x1x1]
%onnx::Conv_775[FLOAT, 32x1x5x5]
%onnx::Conv_778[FLOAT, 32x16x1x1]
%onnx::Conv_781[FLOAT, 32x32x1x1]
%onnx::Conv_784[FLOAT, 32x1x5x5]
%onnx::Conv_787[FLOAT, 32x32x1x1]
%onnx::Conv_790[FLOAT, 32x16x1x1]
%onnx::Conv_793[FLOAT, 32x1x5x5]
%onnx::Conv_796[FLOAT, 64x16x1x1]
%onnx::Conv_797[FLOAT, 64]
%onnx::Conv_799[FLOAT, 384x64x1x1]
%onnx::Conv_800[FLOAT, 384]
%onnx::Conv_802[FLOAT, 384x1x3x3]
%onnx::Conv_805[FLOAT, 64x384x1x1]
%onnx::Conv_808[FLOAT, 64x32x1x1]
%onnx::Conv_811[FLOAT, 64x1x5x5]
%onnx::Conv_814[FLOAT, 64x32x1x1]
%onnx::Conv_817[FLOAT, 64x32x1x1]
%onnx::Conv_820[FLOAT, 64x1x5x5]
%onnx::Conv_823[FLOAT, 64x32x1x1]
%onnx::Conv_826[FLOAT, 192x64x1x1]
%onnx::Conv_829[FLOAT, 192x1x5x5]
%onnx::Conv_832[FLOAT, 112x192x1x1]
%onnx::Conv_833[FLOAT, 112]
%onnx::Conv_835[FLOAT, 672x112x1x1]
%onnx::Conv_836[FLOAT, 672]
%onnx::Conv_838[FLOAT, 672x1x5x5]
%onnx::Conv_841[FLOAT, 112x672x1x1]
%onnx::Conv_844[FLOAT, 672x112x1x1]
%onnx::Conv_847[FLOAT, 672x1x5x5]
%onnx::Conv_850[FLOAT, 112x672x1x1]
%onnx::Conv_853[FLOAT, 112x56x1x1]
%onnx::Conv_856[FLOAT, 112x1x5x5]
%onnx::Conv_859[FLOAT, 112x56x1x1]
%onnx::Conv_862[FLOAT, 672x112x1x1]
%onnx::Conv_865[FLOAT, 672x1x3x3]
%onnx::Conv_868[FLOAT, 184x672x1x1]
%onnx::Conv_869[FLOAT, 184]
%onnx::Conv_871[FLOAT, 184x184x1x1]
%onnx::Conv_874[FLOAT, 184x1x3x3]
%onnx::Conv_877[FLOAT, 184x184x1x1]
%onnx::Conv_880[FLOAT, 184x92x1x1]
%onnx::Conv_883[FLOAT, 184x1x5x5]
%onnx::Conv_886[FLOAT, 184x92x1x1]
%onnx::Conv_889[FLOAT, 1104x184x1x1]
%onnx::Conv_890[FLOAT, 1104]
%onnx::Conv_892[FLOAT, 1104x1x3x3]
%onnx::Conv_895[FLOAT, 352x1104x1x1]
%onnx::Conv_896[FLOAT, 352]
%onnx::Conv_898[FLOAT, 1504x352x1x1]
%onnx::Conv_899[FLOAT, 1504]
) {
%onnx::Conv_893 = Identity(%onnx::Conv_890)
%onnx::Conv_887 = Identity(%onnx::Conv_869)
%onnx::Conv_884 = Identity(%onnx::Conv_869)
%onnx::Conv_881 = Identity(%onnx::Conv_869)
%onnx::Conv_878 = Identity(%onnx::Conv_869)
%onnx::Conv_875 = Identity(%onnx::Conv_869)
%onnx::Conv_872 = Identity(%onnx::Conv_869)
%onnx::Conv_866 = Identity(%onnx::Conv_836)
%onnx::Conv_863 = Identity(%onnx::Conv_836)
%onnx::Conv_860 = Identity(%onnx::Conv_833)
%onnx::Conv_857 = Identity(%onnx::Conv_833)
%onnx::Conv_854 = Identity(%onnx::Conv_833)
%onnx::Conv_851 = Identity(%onnx::Conv_833)
%onnx::Conv_848 = Identity(%onnx::Conv_836)
%onnx::Conv_845 = Identity(%onnx::Conv_836)
%onnx::Conv_842 = Identity(%onnx::Conv_833)
%onnx::Conv_839 = Identity(%onnx::Conv_836)
%onnx::Conv_830 = Identity(%onnx::Conv_764)
%onnx::Conv_827 = Identity(%onnx::Conv_764)
%onnx::Conv_824 = Identity(%onnx::Conv_797)
%onnx::Conv_821 = Identity(%onnx::Conv_797)
%onnx::Conv_818 = Identity(%onnx::Conv_797)
%onnx::Conv_815 = Identity(%onnx::Conv_797)
%onnx::Conv_812 = Identity(%onnx::Conv_797)
%onnx::Conv_809 = Identity(%onnx::Conv_797)
%onnx::Conv_806 = Identity(%onnx::Conv_797)
%onnx::Conv_803 = Identity(%onnx::Conv_800)
%onnx::Conv_794 = Identity(%onnx::Conv_761)
%onnx::Conv_791 = Identity(%onnx::Conv_761)
%onnx::Conv_788 = Identity(%onnx::Conv_761)
%onnx::Conv_785 = Identity(%onnx::Conv_761)
%onnx::Conv_782 = Identity(%onnx::Conv_761)
%onnx::Conv_779 = Identity(%onnx::Conv_761)
%onnx::Conv_776 = Identity(%onnx::Conv_761)
%onnx::Conv_773 = Identity(%onnx::Conv_761)
%onnx::Conv_770 = Identity(%onnx::Conv_761)
%onnx::Conv_767 = Identity(%onnx::Conv_764)
%onnx::Conv_758 = Identity(%onnx::Conv_734)
%onnx::Conv_755 = Identity(%onnx::Conv_734)
%onnx::Conv_752 = Identity(%onnx::Conv_734)
%onnx::Conv_749 = Identity(%onnx::Conv_746)
%onnx::Conv_743 = Identity(%onnx::Conv_734)
%onnx::Conv_740 = Identity(%onnx::Conv_737)
%onnx::Conv_731 = Identity(%onnx::Conv_716)
%onnx::Conv_728 = Identity(%onnx::Conv_716)
%onnx::Conv_725 = Identity(%onnx::Conv_716)
%onnx::Conv_722 = Identity(%onnx::Conv_719)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_715, %onnx::Conv_716)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_862, %onnx::Conv_863)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_865, %onnx::Conv_866)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_868, %onnx::Conv_869)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_871, %onnx::Conv_872)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_874, %onnx::Conv_875)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_877, %onnx::Conv_878)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_880, %onnx::Conv_881)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_output_0 = Reshape(%/cells.20/nl/Relu_output_0, %/cells.20/shuffle/Constant_output_0)
%/cells.20/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.20/shuffle/Reshape_output_0)
%/cells.20/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.20/shuffle/Reshape_1_output_0 = Reshape(%/cells.20/shuffle/Transpose_output_0, %/cells.20/shuffle/Constant_1_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/shuffle/Reshape_1_output_0, %onnx::Conv_883, %onnx::Conv_884)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_886, %onnx::Conv_887)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_889, %onnx::Conv_890)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_892, %onnx::Conv_893)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_895, %onnx::Conv_896)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_898, %onnx::Conv_899)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%713 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %713
}
|
val_accuracy
| 0 | 83,384,448 | 2,116,084 |
{'zcp_synflow': 72.48373293537803, 'zcp_zen': 66.24383544921875, 'zcp_epe_nas': 0.00015999920000638146, 'zcp_fisher': 0.11891056597232819, 'zcp_flops': 83384448.0, 'zcp_grad_norm': 23.920825958251953, 'zcp_grasp': -0.014627456665039062, 'zcp_jacov': -16.0564378798432, 'zcp_l2_norm': 609.665283203125, 'zcp_nwot': 213.99619127684585, 'zcp_params': 2116084.0, 'zcp_plain': -0.002082598628476262, 'zcp_snip': 42.48811340332031, 'lat_1080ti_1': 0.7340788894833844, 'lat_1080ti_32': 0.6103230443338702, 'lat_1080ti_64': 0.5720143291457941, 'lat_2080ti_1': 0.7182917471820496, 'lat_2080ti_32': 0.6617118732150334, 'lat_2080ti_64': 0.6234851765234662, 'lat_essential_ph_1': 0.32075471698113206, 'lat_eyeriss': 0.562906863733886, 'lat_fpga': 0.6296321950543742, 'lat_gold_6226': 0.43661434103869434, 'lat_gold_6240': 0.5697351555054475, 'lat_pixel2': 0.5434782608695652, 'lat_pixel3': 0.6352236469374757, 'lat_raspi4': 0.6525208668865375, 'lat_samsung_a50': 0.2631578947368421, 'lat_samsung_s7': 0.2125984251968504, 'lat_silver_4114': 0.7476893119412223, 'lat_silver_4210r': 0.6384087045704361, 'lat_titan_rtx_1': 0.6544610080121723, 'lat_titan_rtx_32': 0.6261684760304752, 'lat_titan_rtx_64': 0.6277504012921624, 'lat_titanx_1': 0.35370662056600244, 'lat_titanx_32': 0.6221699966387705, 'lat_titanx_64': 0.5370717656612268, 'lat_titanxp_1': 0.6399160980843345, 'lat_titanxp_32': 0.6181291887707678, 'lat_titanxp_64': 0.5892832432927695}
| |
FBNet_1049
|
FBNet
|
1049
|
1049
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_695[FLOAT, 16x3x3x3]
%onnx::Conv_696[FLOAT, 16]
%onnx::Conv_698[FLOAT, 48x16x1x1]
%onnx::Conv_699[FLOAT, 48]
%onnx::Conv_701[FLOAT, 48x1x5x5]
%onnx::Conv_704[FLOAT, 16x48x1x1]
%onnx::Conv_707[FLOAT, 16x16x1x1]
%onnx::Conv_710[FLOAT, 16x1x5x5]
%onnx::Conv_713[FLOAT, 24x16x1x1]
%onnx::Conv_714[FLOAT, 24]
%onnx::Conv_716[FLOAT, 24x12x1x1]
%onnx::Conv_719[FLOAT, 24x1x5x5]
%onnx::Conv_722[FLOAT, 24x12x1x1]
%onnx::Conv_725[FLOAT, 24x24x1x1]
%onnx::Conv_728[FLOAT, 24x1x3x3]
%onnx::Conv_731[FLOAT, 24x24x1x1]
%onnx::Conv_734[FLOAT, 24x24x1x1]
%onnx::Conv_737[FLOAT, 24x1x5x5]
%onnx::Conv_740[FLOAT, 24x24x1x1]
%onnx::Conv_743[FLOAT, 72x24x1x1]
%onnx::Conv_744[FLOAT, 72]
%onnx::Conv_746[FLOAT, 72x1x3x3]
%onnx::Conv_749[FLOAT, 32x72x1x1]
%onnx::Conv_750[FLOAT, 32]
%onnx::Conv_752[FLOAT, 192x32x1x1]
%onnx::Conv_753[FLOAT, 192]
%onnx::Conv_755[FLOAT, 192x1x3x3]
%onnx::Conv_758[FLOAT, 32x192x1x1]
%onnx::Conv_761[FLOAT, 32x32x1x1]
%onnx::Conv_764[FLOAT, 32x1x5x5]
%onnx::Conv_767[FLOAT, 32x32x1x1]
%onnx::Conv_770[FLOAT, 192x32x1x1]
%onnx::Conv_773[FLOAT, 192x1x3x3]
%onnx::Conv_776[FLOAT, 32x192x1x1]
%onnx::Conv_779[FLOAT, 32x16x1x1]
%onnx::Conv_782[FLOAT, 32x1x5x5]
%onnx::Conv_785[FLOAT, 64x16x1x1]
%onnx::Conv_786[FLOAT, 64]
%onnx::Conv_788[FLOAT, 384x64x1x1]
%onnx::Conv_789[FLOAT, 384]
%onnx::Conv_791[FLOAT, 384x1x3x3]
%onnx::Conv_794[FLOAT, 64x384x1x1]
%onnx::Conv_797[FLOAT, 384x64x1x1]
%onnx::Conv_800[FLOAT, 384x1x3x3]
%onnx::Conv_803[FLOAT, 64x384x1x1]
%onnx::Conv_806[FLOAT, 64x32x1x1]
%onnx::Conv_809[FLOAT, 64x1x5x5]
%onnx::Conv_812[FLOAT, 64x32x1x1]
%onnx::Conv_815[FLOAT, 192x64x1x1]
%onnx::Conv_818[FLOAT, 192x1x3x3]
%onnx::Conv_821[FLOAT, 112x192x1x1]
%onnx::Conv_822[FLOAT, 112]
%onnx::Conv_824[FLOAT, 672x112x1x1]
%onnx::Conv_825[FLOAT, 672]
%onnx::Conv_827[FLOAT, 672x1x5x5]
%onnx::Conv_830[FLOAT, 112x672x1x1]
%onnx::Conv_833[FLOAT, 112x112x1x1]
%onnx::Conv_836[FLOAT, 112x1x5x5]
%onnx::Conv_839[FLOAT, 112x112x1x1]
%onnx::Conv_842[FLOAT, 112x56x1x1]
%onnx::Conv_845[FLOAT, 112x1x5x5]
%onnx::Conv_848[FLOAT, 112x56x1x1]
%onnx::Conv_851[FLOAT, 112x112x1x1]
%onnx::Conv_854[FLOAT, 112x1x3x3]
%onnx::Conv_857[FLOAT, 184x112x1x1]
%onnx::Conv_858[FLOAT, 184]
%onnx::Conv_860[FLOAT, 1104x184x1x1]
%onnx::Conv_861[FLOAT, 1104]
%onnx::Conv_863[FLOAT, 1104x1x5x5]
%onnx::Conv_866[FLOAT, 184x1104x1x1]
%onnx::Conv_869[FLOAT, 184x184x1x1]
%onnx::Conv_872[FLOAT, 184x1x5x5]
%onnx::Conv_875[FLOAT, 184x184x1x1]
%onnx::Conv_878[FLOAT, 184x184x1x1]
%onnx::Conv_881[FLOAT, 184x1x3x3]
%onnx::Conv_884[FLOAT, 184x184x1x1]
%onnx::Conv_887[FLOAT, 184x184x1x1]
%onnx::Conv_890[FLOAT, 184x1x5x5]
%onnx::Conv_893[FLOAT, 352x184x1x1]
%onnx::Conv_894[FLOAT, 352]
%onnx::Conv_896[FLOAT, 1504x352x1x1]
%onnx::Conv_897[FLOAT, 1504]
) {
%onnx::Conv_891 = Identity(%onnx::Conv_858)
%onnx::Conv_888 = Identity(%onnx::Conv_858)
%onnx::Conv_885 = Identity(%onnx::Conv_858)
%onnx::Conv_882 = Identity(%onnx::Conv_858)
%onnx::Conv_879 = Identity(%onnx::Conv_858)
%onnx::Conv_876 = Identity(%onnx::Conv_858)
%onnx::Conv_873 = Identity(%onnx::Conv_858)
%onnx::Conv_870 = Identity(%onnx::Conv_858)
%onnx::Conv_867 = Identity(%onnx::Conv_858)
%onnx::Conv_864 = Identity(%onnx::Conv_861)
%onnx::Conv_855 = Identity(%onnx::Conv_822)
%onnx::Conv_852 = Identity(%onnx::Conv_822)
%onnx::Conv_849 = Identity(%onnx::Conv_822)
%onnx::Conv_846 = Identity(%onnx::Conv_822)
%onnx::Conv_843 = Identity(%onnx::Conv_822)
%onnx::Conv_840 = Identity(%onnx::Conv_822)
%onnx::Conv_837 = Identity(%onnx::Conv_822)
%onnx::Conv_834 = Identity(%onnx::Conv_822)
%onnx::Conv_831 = Identity(%onnx::Conv_822)
%onnx::Conv_828 = Identity(%onnx::Conv_825)
%onnx::Conv_819 = Identity(%onnx::Conv_753)
%onnx::Conv_816 = Identity(%onnx::Conv_753)
%onnx::Conv_813 = Identity(%onnx::Conv_786)
%onnx::Conv_810 = Identity(%onnx::Conv_786)
%onnx::Conv_807 = Identity(%onnx::Conv_786)
%onnx::Conv_804 = Identity(%onnx::Conv_786)
%onnx::Conv_801 = Identity(%onnx::Conv_789)
%onnx::Conv_798 = Identity(%onnx::Conv_789)
%onnx::Conv_795 = Identity(%onnx::Conv_786)
%onnx::Conv_792 = Identity(%onnx::Conv_789)
%onnx::Conv_783 = Identity(%onnx::Conv_750)
%onnx::Conv_780 = Identity(%onnx::Conv_750)
%onnx::Conv_777 = Identity(%onnx::Conv_750)
%onnx::Conv_774 = Identity(%onnx::Conv_753)
%onnx::Conv_771 = Identity(%onnx::Conv_753)
%onnx::Conv_768 = Identity(%onnx::Conv_750)
%onnx::Conv_765 = Identity(%onnx::Conv_750)
%onnx::Conv_762 = Identity(%onnx::Conv_750)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_753)
%onnx::Conv_747 = Identity(%onnx::Conv_744)
%onnx::Conv_741 = Identity(%onnx::Conv_714)
%onnx::Conv_738 = Identity(%onnx::Conv_714)
%onnx::Conv_735 = Identity(%onnx::Conv_714)
%onnx::Conv_732 = Identity(%onnx::Conv_714)
%onnx::Conv_729 = Identity(%onnx::Conv_714)
%onnx::Conv_726 = Identity(%onnx::Conv_714)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_714)
%onnx::Conv_717 = Identity(%onnx::Conv_714)
%onnx::Conv_711 = Identity(%onnx::Conv_696)
%onnx::Conv_708 = Identity(%onnx::Conv_696)
%onnx::Conv_705 = Identity(%onnx::Conv_696)
%onnx::Conv_702 = Identity(%onnx::Conv_699)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_695, %onnx::Conv_696)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_821, %onnx::Conv_822)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_824, %onnx::Conv_825)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_827, %onnx::Conv_828)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_830, %onnx::Conv_831)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_833, %onnx::Conv_834)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_836, %onnx::Conv_837)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_839, %onnx::Conv_840)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_842, %onnx::Conv_843)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_845, %onnx::Conv_846)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_848, %onnx::Conv_849)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_851, %onnx::Conv_852)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_854, %onnx::Conv_855)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_857, %onnx::Conv_858)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_860, %onnx::Conv_861)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_863, %onnx::Conv_864)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_866, %onnx::Conv_867)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_869, %onnx::Conv_870)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_872, %onnx::Conv_873)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_875, %onnx::Conv_876)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_878, %onnx::Conv_879)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_881, %onnx::Conv_882)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_884, %onnx::Conv_885)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_887, %onnx::Conv_888)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_890, %onnx::Conv_891)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_893, %onnx::Conv_894)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_896, %onnx::Conv_897)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%693 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %693
}
|
val_accuracy
| 0 | 65,649,792 | 1,820,716 |
{'zcp_synflow': 84.89180702052742, 'zcp_zen': 73.57553100585938, 'zcp_epe_nas': 10.929002565369824, 'zcp_fisher': 0.18323476612567902, 'zcp_flops': 65649792.0, 'zcp_grad_norm': 27.448566436767578, 'zcp_grasp': -0.20920944213867188, 'zcp_jacov': -16.055320375057125, 'zcp_l2_norm': 665.3720703125, 'zcp_nwot': 210.7083278456931, 'zcp_params': 1820716.0, 'zcp_plain': -0.0020065982826054096, 'zcp_snip': 54.24674606323242, 'lat_1080ti_1': 0.6909942548923286, 'lat_1080ti_32': 0.6134216695257598, 'lat_1080ti_64': 0.42755685589260306, 'lat_2080ti_1': 0.7734738829660782, 'lat_2080ti_32': 0.5859107189930503, 'lat_2080ti_64': 0.43240405769159274, 'lat_essential_ph_1': 0.33962264150943394, 'lat_eyeriss': 0.43428749564483893, 'lat_fpga': 0.451261144479957, 'lat_gold_6226': 0.3303558994084615, 'lat_gold_6240': 0.7342319503687842, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.3995073134586372, 'lat_raspi4': 0.41010558392418345, 'lat_samsung_a50': 0.17894736842105263, 'lat_samsung_s7': 0.1732283464566929, 'lat_silver_4114': 0.632000207812093, 'lat_silver_4210r': 0.6789220530619476, 'lat_titan_rtx_1': 0.7524751600727235, 'lat_titan_rtx_32': 0.6108250306913126, 'lat_titan_rtx_64': 0.4686862013603633, 'lat_titanx_1': 0.4010410853484488, 'lat_titanx_32': 0.5275525378171568, 'lat_titanx_64': 0.41738334813476263, 'lat_titanxp_1': 0.7282248264626526, 'lat_titanxp_32': 0.5911366274533832, 'lat_titanxp_64': 0.4378210699763726}
| |
FBNet_4817
|
FBNet
|
4817
|
4817
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_687[FLOAT, 16x3x3x3]
%onnx::Conv_688[FLOAT, 16]
%onnx::Conv_690[FLOAT, 16x8x1x1]
%onnx::Conv_693[FLOAT, 16x1x3x3]
%onnx::Conv_696[FLOAT, 16x8x1x1]
%onnx::Conv_699[FLOAT, 16x16x1x1]
%onnx::Conv_702[FLOAT, 16x1x5x5]
%onnx::Conv_705[FLOAT, 24x16x1x1]
%onnx::Conv_706[FLOAT, 24]
%onnx::Conv_708[FLOAT, 24x12x1x1]
%onnx::Conv_711[FLOAT, 24x1x5x5]
%onnx::Conv_714[FLOAT, 24x12x1x1]
%onnx::Conv_717[FLOAT, 24x24x1x1]
%onnx::Conv_720[FLOAT, 24x1x5x5]
%onnx::Conv_723[FLOAT, 24x24x1x1]
%onnx::Conv_726[FLOAT, 72x24x1x1]
%onnx::Conv_727[FLOAT, 72]
%onnx::Conv_729[FLOAT, 72x1x3x3]
%onnx::Conv_732[FLOAT, 24x72x1x1]
%onnx::Conv_735[FLOAT, 24x24x1x1]
%onnx::Conv_738[FLOAT, 24x1x3x3]
%onnx::Conv_741[FLOAT, 32x24x1x1]
%onnx::Conv_742[FLOAT, 32]
%onnx::Conv_744[FLOAT, 32x32x1x1]
%onnx::Conv_747[FLOAT, 32x1x5x5]
%onnx::Conv_750[FLOAT, 32x32x1x1]
%onnx::Conv_753[FLOAT, 96x32x1x1]
%onnx::Conv_754[FLOAT, 96]
%onnx::Conv_756[FLOAT, 96x1x5x5]
%onnx::Conv_759[FLOAT, 32x96x1x1]
%onnx::Conv_762[FLOAT, 32x32x1x1]
%onnx::Conv_765[FLOAT, 32x1x3x3]
%onnx::Conv_768[FLOAT, 32x32x1x1]
%onnx::Conv_771[FLOAT, 32x16x1x1]
%onnx::Conv_774[FLOAT, 32x1x3x3]
%onnx::Conv_777[FLOAT, 64x16x1x1]
%onnx::Conv_778[FLOAT, 64]
%onnx::Conv_780[FLOAT, 192x64x1x1]
%onnx::Conv_781[FLOAT, 192]
%onnx::Conv_783[FLOAT, 192x1x5x5]
%onnx::Conv_786[FLOAT, 64x192x1x1]
%onnx::Conv_789[FLOAT, 64x32x1x1]
%onnx::Conv_792[FLOAT, 64x1x3x3]
%onnx::Conv_795[FLOAT, 64x32x1x1]
%onnx::Conv_798[FLOAT, 192x64x1x1]
%onnx::Conv_801[FLOAT, 192x1x3x3]
%onnx::Conv_804[FLOAT, 64x192x1x1]
%onnx::Conv_807[FLOAT, 64x64x1x1]
%onnx::Conv_810[FLOAT, 64x1x3x3]
%onnx::Conv_813[FLOAT, 112x64x1x1]
%onnx::Conv_814[FLOAT, 112]
%onnx::Conv_816[FLOAT, 336x112x1x1]
%onnx::Conv_817[FLOAT, 336]
%onnx::Conv_819[FLOAT, 336x1x3x3]
%onnx::Conv_822[FLOAT, 112x336x1x1]
%onnx::Conv_825[FLOAT, 672x112x1x1]
%onnx::Conv_826[FLOAT, 672]
%onnx::Conv_828[FLOAT, 672x1x3x3]
%onnx::Conv_831[FLOAT, 112x672x1x1]
%onnx::Conv_834[FLOAT, 336x112x1x1]
%onnx::Conv_837[FLOAT, 336x1x3x3]
%onnx::Conv_840[FLOAT, 112x336x1x1]
%onnx::Conv_843[FLOAT, 336x112x1x1]
%onnx::Conv_846[FLOAT, 336x1x5x5]
%onnx::Conv_849[FLOAT, 184x336x1x1]
%onnx::Conv_850[FLOAT, 184]
%onnx::Conv_852[FLOAT, 184x184x1x1]
%onnx::Conv_855[FLOAT, 184x1x3x3]
%onnx::Conv_858[FLOAT, 184x184x1x1]
%onnx::Conv_861[FLOAT, 552x184x1x1]
%onnx::Conv_862[FLOAT, 552]
%onnx::Conv_864[FLOAT, 552x1x5x5]
%onnx::Conv_867[FLOAT, 184x552x1x1]
%onnx::Conv_870[FLOAT, 184x92x1x1]
%onnx::Conv_873[FLOAT, 184x1x5x5]
%onnx::Conv_876[FLOAT, 352x92x1x1]
%onnx::Conv_877[FLOAT, 352]
%onnx::Conv_879[FLOAT, 1504x352x1x1]
%onnx::Conv_880[FLOAT, 1504]
) {
%onnx::Conv_874 = Identity(%onnx::Conv_850)
%onnx::Conv_871 = Identity(%onnx::Conv_850)
%onnx::Conv_868 = Identity(%onnx::Conv_850)
%onnx::Conv_865 = Identity(%onnx::Conv_862)
%onnx::Conv_859 = Identity(%onnx::Conv_850)
%onnx::Conv_856 = Identity(%onnx::Conv_850)
%onnx::Conv_853 = Identity(%onnx::Conv_850)
%onnx::Conv_847 = Identity(%onnx::Conv_817)
%onnx::Conv_844 = Identity(%onnx::Conv_817)
%onnx::Conv_841 = Identity(%onnx::Conv_814)
%onnx::Conv_838 = Identity(%onnx::Conv_817)
%onnx::Conv_835 = Identity(%onnx::Conv_817)
%onnx::Conv_832 = Identity(%onnx::Conv_814)
%onnx::Conv_829 = Identity(%onnx::Conv_826)
%onnx::Conv_823 = Identity(%onnx::Conv_814)
%onnx::Conv_820 = Identity(%onnx::Conv_817)
%onnx::Conv_811 = Identity(%onnx::Conv_778)
%onnx::Conv_808 = Identity(%onnx::Conv_778)
%onnx::Conv_805 = Identity(%onnx::Conv_778)
%onnx::Conv_802 = Identity(%onnx::Conv_781)
%onnx::Conv_799 = Identity(%onnx::Conv_781)
%onnx::Conv_796 = Identity(%onnx::Conv_778)
%onnx::Conv_793 = Identity(%onnx::Conv_778)
%onnx::Conv_790 = Identity(%onnx::Conv_778)
%onnx::Conv_787 = Identity(%onnx::Conv_778)
%onnx::Conv_784 = Identity(%onnx::Conv_781)
%onnx::Conv_775 = Identity(%onnx::Conv_742)
%onnx::Conv_772 = Identity(%onnx::Conv_742)
%onnx::Conv_769 = Identity(%onnx::Conv_742)
%onnx::Conv_766 = Identity(%onnx::Conv_742)
%onnx::Conv_763 = Identity(%onnx::Conv_742)
%onnx::Conv_760 = Identity(%onnx::Conv_742)
%onnx::Conv_757 = Identity(%onnx::Conv_754)
%onnx::Conv_751 = Identity(%onnx::Conv_742)
%onnx::Conv_748 = Identity(%onnx::Conv_742)
%onnx::Conv_745 = Identity(%onnx::Conv_742)
%onnx::Conv_739 = Identity(%onnx::Conv_706)
%onnx::Conv_736 = Identity(%onnx::Conv_706)
%onnx::Conv_733 = Identity(%onnx::Conv_706)
%onnx::Conv_730 = Identity(%onnx::Conv_727)
%onnx::Conv_724 = Identity(%onnx::Conv_706)
%onnx::Conv_721 = Identity(%onnx::Conv_706)
%onnx::Conv_718 = Identity(%onnx::Conv_706)
%onnx::Conv_715 = Identity(%onnx::Conv_706)
%onnx::Conv_712 = Identity(%onnx::Conv_706)
%onnx::Conv_709 = Identity(%onnx::Conv_706)
%onnx::Conv_703 = Identity(%onnx::Conv_688)
%onnx::Conv_700 = Identity(%onnx::Conv_688)
%onnx::Conv_697 = Identity(%onnx::Conv_688)
%onnx::Conv_694 = Identity(%onnx::Conv_688)
%onnx::Conv_691 = Identity(%onnx::Conv_688)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_687, %onnx::Conv_688)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_753, %onnx::Conv_754)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_756, %onnx::Conv_757)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_759, %onnx::Conv_760)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_762, %onnx::Conv_763)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_765, %onnx::Conv_766)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_768, %onnx::Conv_769)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_771, %onnx::Conv_772)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_774, %onnx::Conv_775)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_777, %onnx::Conv_778)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_780, %onnx::Conv_781)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_783, %onnx::Conv_784)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_786, %onnx::Conv_787)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_789, %onnx::Conv_790)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_792, %onnx::Conv_793)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_795, %onnx::Conv_796)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_798, %onnx::Conv_799)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_801, %onnx::Conv_802)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_804, %onnx::Conv_805)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_807, %onnx::Conv_808)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_810, %onnx::Conv_811)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_813, %onnx::Conv_814)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_816, %onnx::Conv_817)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_819, %onnx::Conv_820)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_822, %onnx::Conv_823)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_825, %onnx::Conv_826)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_828, %onnx::Conv_829)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_831, %onnx::Conv_832)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_834, %onnx::Conv_835)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_837, %onnx::Conv_838)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_840, %onnx::Conv_841)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_843, %onnx::Conv_844)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_846, %onnx::Conv_847)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_849, %onnx::Conv_850)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_852, %onnx::Conv_853)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_855, %onnx::Conv_856)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_858, %onnx::Conv_859)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_861, %onnx::Conv_862)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_864, %onnx::Conv_865)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_867, %onnx::Conv_868)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_870, %onnx::Conv_871)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_873, %onnx::Conv_874)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_876, %onnx::Conv_877)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_879, %onnx::Conv_880)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%685 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %685
}
|
val_accuracy
| 0 | 58,306,432 | 1,559,996 |
{'zcp_synflow': 78.62211226141828, 'zcp_zen': 67.03286743164062, 'zcp_epe_nas': 6.33637833135894, 'zcp_fisher': 0.14157626032829285, 'zcp_flops': 58306432.0, 'zcp_grad_norm': 22.525714874267578, 'zcp_grasp': 0.07752227783203125, 'zcp_jacov': -16.06952691591237, 'zcp_l2_norm': 594.0748901367188, 'zcp_nwot': 207.37430541216935, 'zcp_params': 1559996.0, 'zcp_plain': 0.005673295818269253, 'zcp_snip': 36.49659729003906, 'lat_1080ti_1': 0.7286909616808714, 'lat_1080ti_32': 0.559041962833184, 'lat_1080ti_64': 0.3045824331555281, 'lat_2080ti_1': 0.7258698557896014, 'lat_2080ti_32': 0.5810264766179343, 'lat_2080ti_64': 0.35478144187095645, 'lat_essential_ph_1': 0.11320754716981132, 'lat_eyeriss': 0.2571195422954908, 'lat_fpga': 0.35525781968948833, 'lat_gold_6226': 0.27322032110926847, 'lat_gold_6240': 0.45434369219774906, 'lat_pixel2': 0.1956521739130435, 'lat_pixel3': 0.28227820239209206, 'lat_raspi4': 0.29748925066556015, 'lat_samsung_a50': 0.12631578947368421, 'lat_samsung_s7': 0.07874015748031496, 'lat_silver_4114': 0.5100562672137035, 'lat_silver_4210r': 0.5245199370520756, 'lat_titan_rtx_1': 0.6797835593485083, 'lat_titan_rtx_32': 0.5676751002608434, 'lat_titan_rtx_64': 0.4070423704841724, 'lat_titanx_1': 0.3622448024193196, 'lat_titanx_32': 0.4588315390334811, 'lat_titanx_64': 0.29443904574885726, 'lat_titanxp_1': 0.6440847192959448, 'lat_titanxp_32': 0.5160217478040092, 'lat_titanxp_64': 0.3447930131003978}
| |
FBNet_2473
|
FBNet
|
2473
|
2473
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_667[FLOAT, 16x3x3x3]
%onnx::Conv_668[FLOAT, 16]
%onnx::Conv_670[FLOAT, 16x8x1x1]
%onnx::Conv_673[FLOAT, 16x1x3x3]
%onnx::Conv_676[FLOAT, 16x8x1x1]
%onnx::Conv_679[FLOAT, 96x16x1x1]
%onnx::Conv_680[FLOAT, 96]
%onnx::Conv_682[FLOAT, 96x1x3x3]
%onnx::Conv_685[FLOAT, 24x96x1x1]
%onnx::Conv_686[FLOAT, 24]
%onnx::Conv_688[FLOAT, 72x24x1x1]
%onnx::Conv_689[FLOAT, 72]
%onnx::Conv_691[FLOAT, 72x1x5x5]
%onnx::Conv_694[FLOAT, 24x72x1x1]
%onnx::Conv_697[FLOAT, 144x24x1x1]
%onnx::Conv_698[FLOAT, 144]
%onnx::Conv_700[FLOAT, 144x1x3x3]
%onnx::Conv_703[FLOAT, 24x144x1x1]
%onnx::Conv_706[FLOAT, 72x24x1x1]
%onnx::Conv_709[FLOAT, 72x1x3x3]
%onnx::Conv_712[FLOAT, 24x72x1x1]
%onnx::Conv_715[FLOAT, 24x24x1x1]
%onnx::Conv_718[FLOAT, 24x1x3x3]
%onnx::Conv_721[FLOAT, 32x24x1x1]
%onnx::Conv_722[FLOAT, 32]
%onnx::Conv_724[FLOAT, 32x32x1x1]
%onnx::Conv_727[FLOAT, 32x1x3x3]
%onnx::Conv_730[FLOAT, 32x32x1x1]
%onnx::Conv_733[FLOAT, 32x32x1x1]
%onnx::Conv_736[FLOAT, 32x1x5x5]
%onnx::Conv_739[FLOAT, 32x32x1x1]
%onnx::Conv_742[FLOAT, 96x32x1x1]
%onnx::Conv_745[FLOAT, 96x1x5x5]
%onnx::Conv_748[FLOAT, 64x96x1x1]
%onnx::Conv_749[FLOAT, 64]
%onnx::Conv_751[FLOAT, 384x64x1x1]
%onnx::Conv_752[FLOAT, 384]
%onnx::Conv_754[FLOAT, 384x1x5x5]
%onnx::Conv_757[FLOAT, 64x384x1x1]
%onnx::Conv_760[FLOAT, 192x64x1x1]
%onnx::Conv_761[FLOAT, 192]
%onnx::Conv_763[FLOAT, 192x1x3x3]
%onnx::Conv_766[FLOAT, 64x192x1x1]
%onnx::Conv_769[FLOAT, 192x64x1x1]
%onnx::Conv_772[FLOAT, 192x1x5x5]
%onnx::Conv_775[FLOAT, 64x192x1x1]
%onnx::Conv_778[FLOAT, 384x64x1x1]
%onnx::Conv_781[FLOAT, 384x1x5x5]
%onnx::Conv_784[FLOAT, 112x384x1x1]
%onnx::Conv_785[FLOAT, 112]
%onnx::Conv_787[FLOAT, 112x112x1x1]
%onnx::Conv_790[FLOAT, 112x1x5x5]
%onnx::Conv_793[FLOAT, 112x112x1x1]
%onnx::Conv_796[FLOAT, 112x56x1x1]
%onnx::Conv_799[FLOAT, 112x1x5x5]
%onnx::Conv_802[FLOAT, 112x56x1x1]
%onnx::Conv_805[FLOAT, 112x56x1x1]
%onnx::Conv_808[FLOAT, 112x1x3x3]
%onnx::Conv_811[FLOAT, 112x56x1x1]
%onnx::Conv_814[FLOAT, 672x112x1x1]
%onnx::Conv_815[FLOAT, 672]
%onnx::Conv_817[FLOAT, 672x1x5x5]
%onnx::Conv_820[FLOAT, 184x672x1x1]
%onnx::Conv_821[FLOAT, 184]
%onnx::Conv_823[FLOAT, 184x92x1x1]
%onnx::Conv_826[FLOAT, 184x1x3x3]
%onnx::Conv_829[FLOAT, 184x92x1x1]
%onnx::Conv_832[FLOAT, 184x184x1x1]
%onnx::Conv_835[FLOAT, 184x1x3x3]
%onnx::Conv_838[FLOAT, 184x184x1x1]
%onnx::Conv_841[FLOAT, 1104x184x1x1]
%onnx::Conv_842[FLOAT, 1104]
%onnx::Conv_844[FLOAT, 1104x1x3x3]
%onnx::Conv_847[FLOAT, 184x1104x1x1]
%onnx::Conv_850[FLOAT, 184x184x1x1]
%onnx::Conv_853[FLOAT, 184x1x3x3]
%onnx::Conv_856[FLOAT, 352x184x1x1]
%onnx::Conv_857[FLOAT, 352]
%onnx::Conv_859[FLOAT, 1504x352x1x1]
%onnx::Conv_860[FLOAT, 1504]
) {
%onnx::Conv_854 = Identity(%onnx::Conv_821)
%onnx::Conv_851 = Identity(%onnx::Conv_821)
%onnx::Conv_848 = Identity(%onnx::Conv_821)
%onnx::Conv_845 = Identity(%onnx::Conv_842)
%onnx::Conv_839 = Identity(%onnx::Conv_821)
%onnx::Conv_836 = Identity(%onnx::Conv_821)
%onnx::Conv_833 = Identity(%onnx::Conv_821)
%onnx::Conv_830 = Identity(%onnx::Conv_821)
%onnx::Conv_827 = Identity(%onnx::Conv_821)
%onnx::Conv_824 = Identity(%onnx::Conv_821)
%onnx::Conv_818 = Identity(%onnx::Conv_815)
%onnx::Conv_812 = Identity(%onnx::Conv_785)
%onnx::Conv_809 = Identity(%onnx::Conv_785)
%onnx::Conv_806 = Identity(%onnx::Conv_785)
%onnx::Conv_803 = Identity(%onnx::Conv_785)
%onnx::Conv_800 = Identity(%onnx::Conv_785)
%onnx::Conv_797 = Identity(%onnx::Conv_785)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_785)
%onnx::Conv_788 = Identity(%onnx::Conv_785)
%onnx::Conv_782 = Identity(%onnx::Conv_752)
%onnx::Conv_779 = Identity(%onnx::Conv_752)
%onnx::Conv_776 = Identity(%onnx::Conv_749)
%onnx::Conv_773 = Identity(%onnx::Conv_761)
%onnx::Conv_770 = Identity(%onnx::Conv_761)
%onnx::Conv_767 = Identity(%onnx::Conv_749)
%onnx::Conv_764 = Identity(%onnx::Conv_761)
%onnx::Conv_758 = Identity(%onnx::Conv_749)
%onnx::Conv_755 = Identity(%onnx::Conv_752)
%onnx::Conv_746 = Identity(%onnx::Conv_680)
%onnx::Conv_743 = Identity(%onnx::Conv_680)
%onnx::Conv_740 = Identity(%onnx::Conv_722)
%onnx::Conv_737 = Identity(%onnx::Conv_722)
%onnx::Conv_734 = Identity(%onnx::Conv_722)
%onnx::Conv_731 = Identity(%onnx::Conv_722)
%onnx::Conv_728 = Identity(%onnx::Conv_722)
%onnx::Conv_725 = Identity(%onnx::Conv_722)
%onnx::Conv_719 = Identity(%onnx::Conv_686)
%onnx::Conv_716 = Identity(%onnx::Conv_686)
%onnx::Conv_713 = Identity(%onnx::Conv_686)
%onnx::Conv_710 = Identity(%onnx::Conv_689)
%onnx::Conv_707 = Identity(%onnx::Conv_689)
%onnx::Conv_704 = Identity(%onnx::Conv_686)
%onnx::Conv_701 = Identity(%onnx::Conv_698)
%onnx::Conv_695 = Identity(%onnx::Conv_686)
%onnx::Conv_692 = Identity(%onnx::Conv_689)
%onnx::Conv_683 = Identity(%onnx::Conv_680)
%onnx::Conv_677 = Identity(%onnx::Conv_668)
%onnx::Conv_674 = Identity(%onnx::Conv_668)
%onnx::Conv_671 = Identity(%onnx::Conv_668)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_667, %onnx::Conv_668)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_output_0 = Reshape(%/cells.15/nl/Relu_output_0, %/cells.15/shuffle/Constant_output_0)
%/cells.15/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.15/shuffle/Reshape_output_0)
%/cells.15/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.15/shuffle/Reshape_1_output_0 = Reshape(%/cells.15/shuffle/Transpose_output_0, %/cells.15/shuffle/Constant_1_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/shuffle/Reshape_1_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_output_0 = Reshape(%/cells.16/nl/Relu_output_0, %/cells.16/shuffle/Constant_output_0)
%/cells.16/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.16/shuffle/Reshape_output_0)
%/cells.16/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.16/shuffle/Reshape_1_output_0 = Reshape(%/cells.16/shuffle/Transpose_output_0, %/cells.16/shuffle/Constant_1_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/shuffle/Reshape_1_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_output_0 = Reshape(%/cells.18/nl/Relu_output_0, %/cells.18/shuffle/Constant_output_0)
%/cells.18/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.18/shuffle/Reshape_output_0)
%/cells.18/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.18/shuffle/Reshape_1_output_0 = Reshape(%/cells.18/shuffle/Transpose_output_0, %/cells.18/shuffle/Constant_1_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.18/shuffle/Reshape_1_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%665 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %665
}
|
val_accuracy
| 0 | 72,225,920 | 1,831,492 |
{'zcp_synflow': 80.8510497218249, 'zcp_zen': 71.32038879394531, 'zcp_epe_nas': 17.571540604286298, 'zcp_fisher': 0.14757999777793884, 'zcp_flops': 72225920.0, 'zcp_grad_norm': 26.940732955932617, 'zcp_grasp': -0.05698966979980469, 'zcp_jacov': -16.061065577234594, 'zcp_l2_norm': 650.298095703125, 'zcp_nwot': 215.1606119915103, 'zcp_params': 1831492.0, 'zcp_plain': 0.00237438827753067, 'zcp_snip': 45.403167724609375, 'lat_1080ti_1': 0.6647796298503245, 'lat_1080ti_32': 0.6635853962015172, 'lat_1080ti_64': 0.5831094728956706, 'lat_2080ti_1': 0.7015567243341746, 'lat_2080ti_32': 0.675941008299941, 'lat_2080ti_64': 0.5981715340704851, 'lat_essential_ph_1': 0.2641509433962264, 'lat_eyeriss': 0.5085544532668295, 'lat_fpga': 0.4946318487220335, 'lat_gold_6226': 0.3466452657116192, 'lat_gold_6240': 0.5160138318816258, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.46705370492438364, 'lat_raspi4': 0.47483568328614345, 'lat_samsung_a50': 0.25263157894736843, 'lat_samsung_s7': 0.1968503937007874, 'lat_silver_4114': 0.5217642354110397, 'lat_silver_4210r': 0.5481959573832467, 'lat_titan_rtx_1': 0.6668612013491172, 'lat_titan_rtx_32': 0.6412231524247333, 'lat_titan_rtx_64': 0.6301877981162818, 'lat_titanx_1': 0.3493855858711709, 'lat_titanx_32': 0.6417718782077784, 'lat_titanx_64': 0.6021850099117643, 'lat_titanxp_1': 0.6100936605821984, 'lat_titanxp_32': 0.6603876423155589, 'lat_titanxp_64': 0.5958759711232556}
| |
FBNet_3541
|
FBNet
|
3541
|
3541
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_634[FLOAT, 16x3x3x3]
%onnx::Conv_635[FLOAT, 16]
%onnx::Conv_637[FLOAT, 48x16x1x1]
%onnx::Conv_638[FLOAT, 48]
%onnx::Conv_640[FLOAT, 48x1x5x5]
%onnx::Conv_643[FLOAT, 16x48x1x1]
%onnx::Conv_646[FLOAT, 16x8x1x1]
%onnx::Conv_649[FLOAT, 16x1x5x5]
%onnx::Conv_652[FLOAT, 24x8x1x1]
%onnx::Conv_653[FLOAT, 24]
%onnx::Conv_655[FLOAT, 24x24x1x1]
%onnx::Conv_658[FLOAT, 24x1x5x5]
%onnx::Conv_661[FLOAT, 24x24x1x1]
%onnx::Conv_664[FLOAT, 144x24x1x1]
%onnx::Conv_665[FLOAT, 144]
%onnx::Conv_667[FLOAT, 144x1x3x3]
%onnx::Conv_670[FLOAT, 24x144x1x1]
%onnx::Conv_673[FLOAT, 24x24x1x1]
%onnx::Conv_676[FLOAT, 24x1x5x5]
%onnx::Conv_679[FLOAT, 24x24x1x1]
%onnx::Conv_682[FLOAT, 24x12x1x1]
%onnx::Conv_685[FLOAT, 24x1x3x3]
%onnx::Conv_688[FLOAT, 32x12x1x1]
%onnx::Conv_689[FLOAT, 32]
%onnx::Conv_691[FLOAT, 32x32x1x1]
%onnx::Conv_694[FLOAT, 32x1x5x5]
%onnx::Conv_697[FLOAT, 32x32x1x1]
%onnx::Conv_700[FLOAT, 192x32x1x1]
%onnx::Conv_701[FLOAT, 192]
%onnx::Conv_703[FLOAT, 192x1x5x5]
%onnx::Conv_706[FLOAT, 32x192x1x1]
%onnx::Conv_709[FLOAT, 192x32x1x1]
%onnx::Conv_712[FLOAT, 192x1x3x3]
%onnx::Conv_715[FLOAT, 32x192x1x1]
%onnx::Conv_718[FLOAT, 32x16x1x1]
%onnx::Conv_721[FLOAT, 32x1x5x5]
%onnx::Conv_724[FLOAT, 64x16x1x1]
%onnx::Conv_725[FLOAT, 64]
%onnx::Conv_727[FLOAT, 192x64x1x1]
%onnx::Conv_730[FLOAT, 192x1x5x5]
%onnx::Conv_733[FLOAT, 64x192x1x1]
%onnx::Conv_736[FLOAT, 192x64x1x1]
%onnx::Conv_739[FLOAT, 192x1x3x3]
%onnx::Conv_742[FLOAT, 64x192x1x1]
%onnx::Conv_745[FLOAT, 384x64x1x1]
%onnx::Conv_746[FLOAT, 384]
%onnx::Conv_748[FLOAT, 384x1x3x3]
%onnx::Conv_751[FLOAT, 64x384x1x1]
%onnx::Conv_754[FLOAT, 112x64x1x1]
%onnx::Conv_755[FLOAT, 112]
%onnx::Conv_757[FLOAT, 112x56x1x1]
%onnx::Conv_760[FLOAT, 112x1x5x5]
%onnx::Conv_763[FLOAT, 112x56x1x1]
%onnx::Conv_766[FLOAT, 672x112x1x1]
%onnx::Conv_767[FLOAT, 672]
%onnx::Conv_769[FLOAT, 672x1x5x5]
%onnx::Conv_772[FLOAT, 112x672x1x1]
%onnx::Conv_775[FLOAT, 112x112x1x1]
%onnx::Conv_778[FLOAT, 112x1x5x5]
%onnx::Conv_781[FLOAT, 184x112x1x1]
%onnx::Conv_782[FLOAT, 184]
%onnx::Conv_784[FLOAT, 552x184x1x1]
%onnx::Conv_785[FLOAT, 552]
%onnx::Conv_787[FLOAT, 552x1x5x5]
%onnx::Conv_790[FLOAT, 184x552x1x1]
%onnx::Conv_793[FLOAT, 184x184x1x1]
%onnx::Conv_796[FLOAT, 184x1x5x5]
%onnx::Conv_799[FLOAT, 184x184x1x1]
%onnx::Conv_802[FLOAT, 552x184x1x1]
%onnx::Conv_805[FLOAT, 552x1x3x3]
%onnx::Conv_808[FLOAT, 184x552x1x1]
%onnx::Conv_811[FLOAT, 352x184x1x1]
%onnx::Conv_812[FLOAT, 352]
%onnx::Conv_814[FLOAT, 1504x352x1x1]
%onnx::Conv_815[FLOAT, 1504]
) {
%onnx::Conv_809 = Identity(%onnx::Conv_782)
%onnx::Conv_806 = Identity(%onnx::Conv_785)
%onnx::Conv_803 = Identity(%onnx::Conv_785)
%onnx::Conv_800 = Identity(%onnx::Conv_782)
%onnx::Conv_797 = Identity(%onnx::Conv_782)
%onnx::Conv_794 = Identity(%onnx::Conv_782)
%onnx::Conv_791 = Identity(%onnx::Conv_782)
%onnx::Conv_788 = Identity(%onnx::Conv_785)
%onnx::Conv_779 = Identity(%onnx::Conv_755)
%onnx::Conv_776 = Identity(%onnx::Conv_755)
%onnx::Conv_773 = Identity(%onnx::Conv_755)
%onnx::Conv_770 = Identity(%onnx::Conv_767)
%onnx::Conv_764 = Identity(%onnx::Conv_755)
%onnx::Conv_761 = Identity(%onnx::Conv_755)
%onnx::Conv_758 = Identity(%onnx::Conv_755)
%onnx::Conv_752 = Identity(%onnx::Conv_725)
%onnx::Conv_749 = Identity(%onnx::Conv_746)
%onnx::Conv_743 = Identity(%onnx::Conv_725)
%onnx::Conv_740 = Identity(%onnx::Conv_701)
%onnx::Conv_737 = Identity(%onnx::Conv_701)
%onnx::Conv_734 = Identity(%onnx::Conv_725)
%onnx::Conv_731 = Identity(%onnx::Conv_701)
%onnx::Conv_728 = Identity(%onnx::Conv_701)
%onnx::Conv_722 = Identity(%onnx::Conv_689)
%onnx::Conv_719 = Identity(%onnx::Conv_689)
%onnx::Conv_716 = Identity(%onnx::Conv_689)
%onnx::Conv_713 = Identity(%onnx::Conv_701)
%onnx::Conv_710 = Identity(%onnx::Conv_701)
%onnx::Conv_707 = Identity(%onnx::Conv_689)
%onnx::Conv_704 = Identity(%onnx::Conv_701)
%onnx::Conv_698 = Identity(%onnx::Conv_689)
%onnx::Conv_695 = Identity(%onnx::Conv_689)
%onnx::Conv_692 = Identity(%onnx::Conv_689)
%onnx::Conv_686 = Identity(%onnx::Conv_653)
%onnx::Conv_683 = Identity(%onnx::Conv_653)
%onnx::Conv_680 = Identity(%onnx::Conv_653)
%onnx::Conv_677 = Identity(%onnx::Conv_653)
%onnx::Conv_674 = Identity(%onnx::Conv_653)
%onnx::Conv_671 = Identity(%onnx::Conv_653)
%onnx::Conv_668 = Identity(%onnx::Conv_665)
%onnx::Conv_662 = Identity(%onnx::Conv_653)
%onnx::Conv_659 = Identity(%onnx::Conv_653)
%onnx::Conv_656 = Identity(%onnx::Conv_653)
%onnx::Conv_650 = Identity(%onnx::Conv_635)
%onnx::Conv_647 = Identity(%onnx::Conv_635)
%onnx::Conv_644 = Identity(%onnx::Conv_635)
%onnx::Conv_641 = Identity(%onnx::Conv_638)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_634, %onnx::Conv_635)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_637, %onnx::Conv_638)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_640, %onnx::Conv_641)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_643, %onnx::Conv_644)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_646, %onnx::Conv_647)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_649, %onnx::Conv_650)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_652, %onnx::Conv_653)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_655, %onnx::Conv_656)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_658, %onnx::Conv_659)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_661, %onnx::Conv_662)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_664, %onnx::Conv_665)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_667, %onnx::Conv_668)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_670, %onnx::Conv_671)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_673, %onnx::Conv_674)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_676, %onnx::Conv_677)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.13/relu/Relu_output_0 = Relu(%/cells.13/conv/Conv_output_0)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/relu/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_output_0 = Reshape(%/cells.14/nl/Relu_output_0, %/cells.14/shuffle/Constant_output_0)
%/cells.14/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.14/shuffle/Reshape_output_0)
%/cells.14/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.14/shuffle/Reshape_1_output_0 = Reshape(%/cells.14/shuffle/Transpose_output_0, %/cells.14/shuffle/Constant_1_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/shuffle/Reshape_1_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/relu/Relu_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.21/relu/Relu_output_0 = Relu(%/cells.21/conv/Conv_output_0)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/relu/Relu_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%632 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %632
}
|
val_accuracy
| 0 | 66,351,232 | 1,650,484 |
{'zcp_synflow': 78.76030489185978, 'zcp_zen': 66.72696685791016, 'zcp_epe_nas': 9.362565043092344, 'zcp_fisher': 0.09451635181903839, 'zcp_flops': 66351232.0, 'zcp_grad_norm': 22.298614501953125, 'zcp_grasp': 0.017002105712890625, 'zcp_jacov': -16.04998550019865, 'zcp_l2_norm': 599.904296875, 'zcp_nwot': 212.78350360924705, 'zcp_params': 1650484.0, 'zcp_plain': 0.009090427309274673, 'zcp_snip': 36.67967224121094, 'lat_1080ti_1': 0.5867368297907117, 'lat_1080ti_32': 0.5449472446952982, 'lat_1080ti_64': 0.46356531652315347, 'lat_2080ti_1': 0.5855004853408845, 'lat_2080ti_32': 0.5080953288927813, 'lat_2080ti_64': 0.4842038209110163, 'lat_essential_ph_1': 0.2830188679245283, 'lat_eyeriss': 0.44107237819302064, 'lat_fpga': 0.43426117416558624, 'lat_gold_6226': 0.3178717882421802, 'lat_gold_6240': 0.5000245294358813, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.4412471204788462, 'lat_raspi4': 0.4340663440245313, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.14960629921259844, 'lat_silver_4114': 0.5202917677881062, 'lat_silver_4210r': 0.521309276625203, 'lat_titan_rtx_1': 0.5522758375818426, 'lat_titan_rtx_32': 0.4967672462624751, 'lat_titan_rtx_64': 0.48733075453269364, 'lat_titanx_1': 0.2906906787407627, 'lat_titanx_32': 0.4782405544327084, 'lat_titanx_64': 0.44821597933194735, 'lat_titanxp_1': 0.5303545765221169, 'lat_titanxp_32': 0.48030044560725765, 'lat_titanxp_64': 0.4702479808274593}
| |
FBNet_4851
|
FBNet
|
4851
|
4851
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_641[FLOAT, 16x3x3x3]
%onnx::Conv_642[FLOAT, 16]
%onnx::Conv_644[FLOAT, 16x16x1x1]
%onnx::Conv_647[FLOAT, 16x1x3x3]
%onnx::Conv_650[FLOAT, 24x16x1x1]
%onnx::Conv_651[FLOAT, 24]
%onnx::Conv_653[FLOAT, 24x12x1x1]
%onnx::Conv_656[FLOAT, 24x1x3x3]
%onnx::Conv_659[FLOAT, 24x12x1x1]
%onnx::Conv_662[FLOAT, 24x24x1x1]
%onnx::Conv_665[FLOAT, 24x1x3x3]
%onnx::Conv_668[FLOAT, 24x24x1x1]
%onnx::Conv_671[FLOAT, 144x24x1x1]
%onnx::Conv_672[FLOAT, 144]
%onnx::Conv_674[FLOAT, 144x1x3x3]
%onnx::Conv_677[FLOAT, 32x144x1x1]
%onnx::Conv_678[FLOAT, 32]
%onnx::Conv_680[FLOAT, 96x32x1x1]
%onnx::Conv_681[FLOAT, 96]
%onnx::Conv_683[FLOAT, 96x1x5x5]
%onnx::Conv_686[FLOAT, 32x96x1x1]
%onnx::Conv_689[FLOAT, 96x32x1x1]
%onnx::Conv_692[FLOAT, 96x1x3x3]
%onnx::Conv_695[FLOAT, 32x96x1x1]
%onnx::Conv_698[FLOAT, 32x32x1x1]
%onnx::Conv_701[FLOAT, 32x1x5x5]
%onnx::Conv_704[FLOAT, 32x32x1x1]
%onnx::Conv_707[FLOAT, 32x16x1x1]
%onnx::Conv_710[FLOAT, 32x1x5x5]
%onnx::Conv_713[FLOAT, 64x16x1x1]
%onnx::Conv_714[FLOAT, 64]
%onnx::Conv_716[FLOAT, 192x64x1x1]
%onnx::Conv_717[FLOAT, 192]
%onnx::Conv_719[FLOAT, 192x1x5x5]
%onnx::Conv_722[FLOAT, 64x192x1x1]
%onnx::Conv_725[FLOAT, 64x32x1x1]
%onnx::Conv_728[FLOAT, 64x1x5x5]
%onnx::Conv_731[FLOAT, 64x32x1x1]
%onnx::Conv_734[FLOAT, 64x32x1x1]
%onnx::Conv_737[FLOAT, 64x1x5x5]
%onnx::Conv_740[FLOAT, 64x32x1x1]
%onnx::Conv_743[FLOAT, 64x32x1x1]
%onnx::Conv_746[FLOAT, 64x1x3x3]
%onnx::Conv_749[FLOAT, 112x32x1x1]
%onnx::Conv_750[FLOAT, 112]
%onnx::Conv_752[FLOAT, 112x112x1x1]
%onnx::Conv_755[FLOAT, 112x1x5x5]
%onnx::Conv_758[FLOAT, 112x112x1x1]
%onnx::Conv_761[FLOAT, 112x112x1x1]
%onnx::Conv_764[FLOAT, 112x1x3x3]
%onnx::Conv_767[FLOAT, 112x112x1x1]
%onnx::Conv_770[FLOAT, 336x112x1x1]
%onnx::Conv_771[FLOAT, 336]
%onnx::Conv_773[FLOAT, 336x1x3x3]
%onnx::Conv_776[FLOAT, 112x336x1x1]
%onnx::Conv_779[FLOAT, 184x112x1x1]
%onnx::Conv_780[FLOAT, 184]
%onnx::Conv_782[FLOAT, 552x184x1x1]
%onnx::Conv_783[FLOAT, 552]
%onnx::Conv_785[FLOAT, 552x1x5x5]
%onnx::Conv_788[FLOAT, 184x552x1x1]
%onnx::Conv_791[FLOAT, 552x184x1x1]
%onnx::Conv_794[FLOAT, 552x1x5x5]
%onnx::Conv_797[FLOAT, 184x552x1x1]
%onnx::Conv_800[FLOAT, 552x184x1x1]
%onnx::Conv_803[FLOAT, 552x1x5x5]
%onnx::Conv_806[FLOAT, 184x552x1x1]
%onnx::Conv_809[FLOAT, 1104x184x1x1]
%onnx::Conv_810[FLOAT, 1104]
%onnx::Conv_812[FLOAT, 1104x1x3x3]
%onnx::Conv_815[FLOAT, 352x1104x1x1]
%onnx::Conv_816[FLOAT, 352]
%onnx::Conv_818[FLOAT, 1504x352x1x1]
%onnx::Conv_819[FLOAT, 1504]
) {
%onnx::Conv_813 = Identity(%onnx::Conv_810)
%onnx::Conv_807 = Identity(%onnx::Conv_780)
%onnx::Conv_804 = Identity(%onnx::Conv_783)
%onnx::Conv_801 = Identity(%onnx::Conv_783)
%onnx::Conv_798 = Identity(%onnx::Conv_780)
%onnx::Conv_795 = Identity(%onnx::Conv_783)
%onnx::Conv_792 = Identity(%onnx::Conv_783)
%onnx::Conv_789 = Identity(%onnx::Conv_780)
%onnx::Conv_786 = Identity(%onnx::Conv_783)
%onnx::Conv_777 = Identity(%onnx::Conv_750)
%onnx::Conv_774 = Identity(%onnx::Conv_771)
%onnx::Conv_768 = Identity(%onnx::Conv_750)
%onnx::Conv_765 = Identity(%onnx::Conv_750)
%onnx::Conv_762 = Identity(%onnx::Conv_750)
%onnx::Conv_759 = Identity(%onnx::Conv_750)
%onnx::Conv_756 = Identity(%onnx::Conv_750)
%onnx::Conv_753 = Identity(%onnx::Conv_750)
%onnx::Conv_747 = Identity(%onnx::Conv_714)
%onnx::Conv_744 = Identity(%onnx::Conv_714)
%onnx::Conv_741 = Identity(%onnx::Conv_714)
%onnx::Conv_738 = Identity(%onnx::Conv_714)
%onnx::Conv_735 = Identity(%onnx::Conv_714)
%onnx::Conv_732 = Identity(%onnx::Conv_714)
%onnx::Conv_729 = Identity(%onnx::Conv_714)
%onnx::Conv_726 = Identity(%onnx::Conv_714)
%onnx::Conv_723 = Identity(%onnx::Conv_714)
%onnx::Conv_720 = Identity(%onnx::Conv_717)
%onnx::Conv_711 = Identity(%onnx::Conv_678)
%onnx::Conv_708 = Identity(%onnx::Conv_678)
%onnx::Conv_705 = Identity(%onnx::Conv_678)
%onnx::Conv_702 = Identity(%onnx::Conv_678)
%onnx::Conv_699 = Identity(%onnx::Conv_678)
%onnx::Conv_696 = Identity(%onnx::Conv_678)
%onnx::Conv_693 = Identity(%onnx::Conv_681)
%onnx::Conv_690 = Identity(%onnx::Conv_681)
%onnx::Conv_687 = Identity(%onnx::Conv_678)
%onnx::Conv_684 = Identity(%onnx::Conv_681)
%onnx::Conv_675 = Identity(%onnx::Conv_672)
%onnx::Conv_669 = Identity(%onnx::Conv_651)
%onnx::Conv_666 = Identity(%onnx::Conv_651)
%onnx::Conv_663 = Identity(%onnx::Conv_651)
%onnx::Conv_660 = Identity(%onnx::Conv_651)
%onnx::Conv_657 = Identity(%onnx::Conv_651)
%onnx::Conv_654 = Identity(%onnx::Conv_651)
%onnx::Conv_648 = Identity(%onnx::Conv_642)
%onnx::Conv_645 = Identity(%onnx::Conv_642)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_641, %onnx::Conv_642)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_644, %onnx::Conv_645)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_647, %onnx::Conv_648)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_650, %onnx::Conv_651)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_653, %onnx::Conv_654)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_output_0 = Reshape(%/cells.2/nl/Relu_output_0, %/cells.2/shuffle/Constant_output_0)
%/cells.2/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.2/shuffle/Reshape_output_0)
%/cells.2/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.2/shuffle/Reshape_1_output_0 = Reshape(%/cells.2/shuffle/Transpose_output_0, %/cells.2/shuffle/Constant_1_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/shuffle/Reshape_1_output_0, %onnx::Conv_656, %onnx::Conv_657)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_659, %onnx::Conv_660)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_662, %onnx::Conv_663)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_665, %onnx::Conv_666)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_668, %onnx::Conv_669)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_671, %onnx::Conv_672)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_674, %onnx::Conv_675)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_677, %onnx::Conv_678)
%/cells.6/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_680, %onnx::Conv_681)
%/cells.6/nl/Relu_output_0 = Relu(%/cells.6/conv1/Conv_output_0)
%/cells.6/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.6/nl/Relu_output_0, %onnx::Conv_683, %onnx::Conv_684)
%/cells.6/nl_1/Relu_output_0 = Relu(%/cells.6/conv2/Conv_output_0)
%/cells.6/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/nl_1/Relu_output_0, %onnx::Conv_686, %onnx::Conv_687)
%/cells.6/Add_output_0 = Add(%/cells.6/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.6/Add_output_0, %onnx::Conv_689, %onnx::Conv_690)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_692, %onnx::Conv_693)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_695, %onnx::Conv_696)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.6/Add_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_698, %onnx::Conv_699)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_701, %onnx::Conv_702)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_704, %onnx::Conv_705)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_707, %onnx::Conv_708)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_output_0 = Reshape(%/cells.9/nl/Relu_output_0, %/cells.9/shuffle/Constant_output_0)
%/cells.9/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.9/shuffle/Reshape_output_0)
%/cells.9/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.9/shuffle/Reshape_1_output_0 = Reshape(%/cells.9/shuffle/Transpose_output_0, %/cells.9/shuffle/Constant_1_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/shuffle/Reshape_1_output_0, %onnx::Conv_710, %onnx::Conv_711)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_713, %onnx::Conv_714)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_716, %onnx::Conv_717)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_719, %onnx::Conv_720)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_722, %onnx::Conv_723)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_725, %onnx::Conv_726)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_output_0 = Reshape(%/cells.11/nl/Relu_output_0, %/cells.11/shuffle/Constant_output_0)
%/cells.11/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.11/shuffle/Reshape_output_0)
%/cells.11/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.11/shuffle/Reshape_1_output_0 = Reshape(%/cells.11/shuffle/Transpose_output_0, %/cells.11/shuffle/Constant_1_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.11/shuffle/Reshape_1_output_0, %onnx::Conv_728, %onnx::Conv_729)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_731, %onnx::Conv_732)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_734, %onnx::Conv_735)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_output_0 = Reshape(%/cells.12/nl/Relu_output_0, %/cells.12/shuffle/Constant_output_0)
%/cells.12/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.12/shuffle/Reshape_output_0)
%/cells.12/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.12/shuffle/Reshape_1_output_0 = Reshape(%/cells.12/shuffle/Transpose_output_0, %/cells.12/shuffle/Constant_1_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/shuffle/Reshape_1_output_0, %onnx::Conv_737, %onnx::Conv_738)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_740, %onnx::Conv_741)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_743, %onnx::Conv_744)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_output_0 = Reshape(%/cells.13/nl/Relu_output_0, %/cells.13/shuffle/Constant_output_0)
%/cells.13/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.13/shuffle/Reshape_output_0)
%/cells.13/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.13/shuffle/Reshape_1_output_0 = Reshape(%/cells.13/shuffle/Transpose_output_0, %/cells.13/shuffle/Constant_1_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/shuffle/Reshape_1_output_0, %onnx::Conv_746, %onnx::Conv_747)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_749, %onnx::Conv_750)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_752, %onnx::Conv_753)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_755, %onnx::Conv_756)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_758, %onnx::Conv_759)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_761, %onnx::Conv_762)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_764, %onnx::Conv_765)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_767, %onnx::Conv_768)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_770, %onnx::Conv_771)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_773, %onnx::Conv_774)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_776, %onnx::Conv_777)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [2, 2]](%/cells.16/Add_output_0, %onnx::Conv_779, %onnx::Conv_780)
%/cells.17/relu/Relu_output_0 = Relu(%/cells.17/conv/Conv_output_0)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/relu/Relu_output_0, %onnx::Conv_782, %onnx::Conv_783)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_785, %onnx::Conv_786)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_788, %onnx::Conv_789)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/relu/Relu_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_791, %onnx::Conv_792)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_794, %onnx::Conv_795)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_797, %onnx::Conv_798)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_800, %onnx::Conv_801)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_803, %onnx::Conv_804)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_806, %onnx::Conv_807)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_809, %onnx::Conv_810)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_812, %onnx::Conv_813)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_815, %onnx::Conv_816)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_818, %onnx::Conv_819)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%639 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %639
}
|
val_accuracy
| 0 | 55,979,520 | 2,189,420 |
{'zcp_synflow': 74.69438148186907, 'zcp_zen': 65.67754364013672, 'zcp_epe_nas': 5.192194401729585, 'zcp_fisher': 0.06269396096467972, 'zcp_flops': 55979520.0, 'zcp_grad_norm': 18.030277252197266, 'zcp_grasp': -0.03856849670410156, 'zcp_jacov': -16.11716986462006, 'zcp_l2_norm': 604.5126953125, 'zcp_nwot': 205.62016647919918, 'zcp_params': 2189420.0, 'zcp_plain': -0.00039960560388863087, 'zcp_snip': 31.30953025817871, 'lat_1080ti_1': 0.5861169444045576, 'lat_1080ti_32': 0.39862642055460573, 'lat_1080ti_64': 0.21675255830246978, 'lat_2080ti_1': 0.5665423254377114, 'lat_2080ti_32': 0.4574024509801324, 'lat_2080ti_64': 0.2765232078358555, 'lat_essential_ph_1': 0.11320754716981132, 'lat_eyeriss': 0.2680303669337829, 'lat_fpga': 0.33118277441890365, 'lat_gold_6226': 0.3267748826648286, 'lat_gold_6240': 0.49296356301668537, 'lat_pixel2': 0.2608695652173913, 'lat_pixel3': 0.24634775322126976, 'lat_raspi4': 0.3285124229218231, 'lat_samsung_a50': 0.1368421052631579, 'lat_samsung_s7': 0.14173228346456693, 'lat_silver_4114': 0.4328065014219418, 'lat_silver_4210r': 0.4229434881617467, 'lat_titan_rtx_1': 0.5401362420588283, 'lat_titan_rtx_32': 0.4391404550297319, 'lat_titan_rtx_64': 0.3038090089008292, 'lat_titanx_1': 0.2906797530064099, 'lat_titanx_32': 0.337561349057824, 'lat_titanx_64': 0.20929295533495956, 'lat_titanxp_1': 0.5132988885746051, 'lat_titanxp_32': 0.3810782551358478, 'lat_titanxp_64': 0.21546829251465108}
| |
FBNet_2914
|
FBNet
|
2914
|
2914
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_576[FLOAT, 16x3x3x3]
%onnx::Conv_577[FLOAT, 16]
%onnx::Conv_579[FLOAT, 48x16x1x1]
%onnx::Conv_580[FLOAT, 48]
%onnx::Conv_582[FLOAT, 48x1x5x5]
%onnx::Conv_585[FLOAT, 16x48x1x1]
%onnx::Conv_588[FLOAT, 48x16x1x1]
%onnx::Conv_591[FLOAT, 48x1x3x3]
%onnx::Conv_594[FLOAT, 24x48x1x1]
%onnx::Conv_595[FLOAT, 24]
%onnx::Conv_597[FLOAT, 72x24x1x1]
%onnx::Conv_598[FLOAT, 72]
%onnx::Conv_600[FLOAT, 72x1x5x5]
%onnx::Conv_603[FLOAT, 24x72x1x1]
%onnx::Conv_606[FLOAT, 24x24x1x1]
%onnx::Conv_609[FLOAT, 24x1x3x3]
%onnx::Conv_612[FLOAT, 24x24x1x1]
%onnx::Conv_615[FLOAT, 72x24x1x1]
%onnx::Conv_618[FLOAT, 72x1x5x5]
%onnx::Conv_621[FLOAT, 32x72x1x1]
%onnx::Conv_622[FLOAT, 32]
%onnx::Conv_624[FLOAT, 32x16x1x1]
%onnx::Conv_627[FLOAT, 32x1x5x5]
%onnx::Conv_630[FLOAT, 32x16x1x1]
%onnx::Conv_633[FLOAT, 96x32x1x1]
%onnx::Conv_634[FLOAT, 96]
%onnx::Conv_636[FLOAT, 96x1x5x5]
%onnx::Conv_639[FLOAT, 32x96x1x1]
%onnx::Conv_642[FLOAT, 192x32x1x1]
%onnx::Conv_643[FLOAT, 192]
%onnx::Conv_645[FLOAT, 192x1x5x5]
%onnx::Conv_648[FLOAT, 64x192x1x1]
%onnx::Conv_649[FLOAT, 64]
%onnx::Conv_651[FLOAT, 64x32x1x1]
%onnx::Conv_654[FLOAT, 64x1x5x5]
%onnx::Conv_657[FLOAT, 64x32x1x1]
%onnx::Conv_660[FLOAT, 384x64x1x1]
%onnx::Conv_661[FLOAT, 384]
%onnx::Conv_663[FLOAT, 384x1x3x3]
%onnx::Conv_666[FLOAT, 64x384x1x1]
%onnx::Conv_669[FLOAT, 384x64x1x1]
%onnx::Conv_672[FLOAT, 384x1x5x5]
%onnx::Conv_675[FLOAT, 64x384x1x1]
%onnx::Conv_678[FLOAT, 64x64x1x1]
%onnx::Conv_681[FLOAT, 64x1x3x3]
%onnx::Conv_684[FLOAT, 112x64x1x1]
%onnx::Conv_685[FLOAT, 112]
%onnx::Conv_687[FLOAT, 672x112x1x1]
%onnx::Conv_688[FLOAT, 672]
%onnx::Conv_690[FLOAT, 672x1x3x3]
%onnx::Conv_693[FLOAT, 112x672x1x1]
%onnx::Conv_696[FLOAT, 336x112x1x1]
%onnx::Conv_697[FLOAT, 336]
%onnx::Conv_699[FLOAT, 336x1x5x5]
%onnx::Conv_702[FLOAT, 112x336x1x1]
%onnx::Conv_705[FLOAT, 672x112x1x1]
%onnx::Conv_708[FLOAT, 672x1x5x5]
%onnx::Conv_711[FLOAT, 184x672x1x1]
%onnx::Conv_712[FLOAT, 184]
%onnx::Conv_714[FLOAT, 184x184x1x1]
%onnx::Conv_717[FLOAT, 184x1x5x5]
%onnx::Conv_720[FLOAT, 184x184x1x1]
%onnx::Conv_723[FLOAT, 552x184x1x1]
%onnx::Conv_724[FLOAT, 552]
%onnx::Conv_726[FLOAT, 552x1x5x5]
%onnx::Conv_729[FLOAT, 184x552x1x1]
%onnx::Conv_732[FLOAT, 1104x184x1x1]
%onnx::Conv_733[FLOAT, 1104]
%onnx::Conv_735[FLOAT, 1104x1x5x5]
%onnx::Conv_738[FLOAT, 184x1104x1x1]
%onnx::Conv_741[FLOAT, 1104x184x1x1]
%onnx::Conv_744[FLOAT, 1104x1x3x3]
%onnx::Conv_747[FLOAT, 352x1104x1x1]
%onnx::Conv_748[FLOAT, 352]
%onnx::Conv_750[FLOAT, 1504x352x1x1]
%onnx::Conv_751[FLOAT, 1504]
) {
%onnx::Conv_745 = Identity(%onnx::Conv_733)
%onnx::Conv_742 = Identity(%onnx::Conv_733)
%onnx::Conv_739 = Identity(%onnx::Conv_712)
%onnx::Conv_736 = Identity(%onnx::Conv_733)
%onnx::Conv_730 = Identity(%onnx::Conv_712)
%onnx::Conv_727 = Identity(%onnx::Conv_724)
%onnx::Conv_721 = Identity(%onnx::Conv_712)
%onnx::Conv_718 = Identity(%onnx::Conv_712)
%onnx::Conv_715 = Identity(%onnx::Conv_712)
%onnx::Conv_709 = Identity(%onnx::Conv_688)
%onnx::Conv_706 = Identity(%onnx::Conv_688)
%onnx::Conv_703 = Identity(%onnx::Conv_685)
%onnx::Conv_700 = Identity(%onnx::Conv_697)
%onnx::Conv_694 = Identity(%onnx::Conv_685)
%onnx::Conv_691 = Identity(%onnx::Conv_688)
%onnx::Conv_682 = Identity(%onnx::Conv_649)
%onnx::Conv_679 = Identity(%onnx::Conv_649)
%onnx::Conv_676 = Identity(%onnx::Conv_649)
%onnx::Conv_673 = Identity(%onnx::Conv_661)
%onnx::Conv_670 = Identity(%onnx::Conv_661)
%onnx::Conv_667 = Identity(%onnx::Conv_649)
%onnx::Conv_664 = Identity(%onnx::Conv_661)
%onnx::Conv_658 = Identity(%onnx::Conv_649)
%onnx::Conv_655 = Identity(%onnx::Conv_649)
%onnx::Conv_652 = Identity(%onnx::Conv_649)
%onnx::Conv_646 = Identity(%onnx::Conv_643)
%onnx::Conv_640 = Identity(%onnx::Conv_622)
%onnx::Conv_637 = Identity(%onnx::Conv_634)
%onnx::Conv_631 = Identity(%onnx::Conv_622)
%onnx::Conv_628 = Identity(%onnx::Conv_622)
%onnx::Conv_625 = Identity(%onnx::Conv_622)
%onnx::Conv_619 = Identity(%onnx::Conv_598)
%onnx::Conv_616 = Identity(%onnx::Conv_598)
%onnx::Conv_613 = Identity(%onnx::Conv_595)
%onnx::Conv_610 = Identity(%onnx::Conv_595)
%onnx::Conv_607 = Identity(%onnx::Conv_595)
%onnx::Conv_604 = Identity(%onnx::Conv_595)
%onnx::Conv_601 = Identity(%onnx::Conv_598)
%onnx::Conv_592 = Identity(%onnx::Conv_580)
%onnx::Conv_589 = Identity(%onnx::Conv_580)
%onnx::Conv_586 = Identity(%onnx::Conv_577)
%onnx::Conv_583 = Identity(%onnx::Conv_580)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_576, %onnx::Conv_577)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_579, %onnx::Conv_580)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.0/nl/Relu_output_0, %onnx::Conv_582, %onnx::Conv_583)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_585, %onnx::Conv_586)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_588, %onnx::Conv_589)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 48, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/nl/Relu_output_0, %onnx::Conv_591, %onnx::Conv_592)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_594, %onnx::Conv_595)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_597, %onnx::Conv_598)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_600, %onnx::Conv_601)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_603, %onnx::Conv_604)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_606, %onnx::Conv_607)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/nl/Relu_output_0, %onnx::Conv_609, %onnx::Conv_610)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_612, %onnx::Conv_613)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_615, %onnx::Conv_616)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.5/nl/Relu_output_0, %onnx::Conv_618, %onnx::Conv_619)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_621, %onnx::Conv_622)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_624, %onnx::Conv_625)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_output_0 = Reshape(%/cells.7/nl/Relu_output_0, %/cells.7/shuffle/Constant_output_0)
%/cells.7/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.7/shuffle/Reshape_output_0)
%/cells.7/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.7/shuffle/Reshape_1_output_0 = Reshape(%/cells.7/shuffle/Transpose_output_0, %/cells.7/shuffle/Constant_1_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.7/shuffle/Reshape_1_output_0, %onnx::Conv_627, %onnx::Conv_628)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_630, %onnx::Conv_631)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_633, %onnx::Conv_634)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_636, %onnx::Conv_637)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_639, %onnx::Conv_640)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_642, %onnx::Conv_643)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_645, %onnx::Conv_646)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_648, %onnx::Conv_649)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_651, %onnx::Conv_652)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_output_0 = Reshape(%/cells.10/nl/Relu_output_0, %/cells.10/shuffle/Constant_output_0)
%/cells.10/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.10/shuffle/Reshape_output_0)
%/cells.10/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.10/shuffle/Reshape_1_output_0 = Reshape(%/cells.10/shuffle/Transpose_output_0, %/cells.10/shuffle/Constant_1_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/shuffle/Reshape_1_output_0, %onnx::Conv_654, %onnx::Conv_655)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_657, %onnx::Conv_658)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.11/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_660, %onnx::Conv_661)
%/cells.11/nl/Relu_output_0 = Relu(%/cells.11/conv1/Conv_output_0)
%/cells.11/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.11/nl/Relu_output_0, %onnx::Conv_663, %onnx::Conv_664)
%/cells.11/nl_1/Relu_output_0 = Relu(%/cells.11/conv2/Conv_output_0)
%/cells.11/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/nl_1/Relu_output_0, %onnx::Conv_666, %onnx::Conv_667)
%/cells.11/Add_output_0 = Add(%/cells.11/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.11/Add_output_0, %onnx::Conv_669, %onnx::Conv_670)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 384, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_672, %onnx::Conv_673)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_675, %onnx::Conv_676)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.11/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_678, %onnx::Conv_679)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_681, %onnx::Conv_682)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_684, %onnx::Conv_685)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_687, %onnx::Conv_688)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_690, %onnx::Conv_691)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_693, %onnx::Conv_694)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_696, %onnx::Conv_697)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 336, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_699, %onnx::Conv_700)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_702, %onnx::Conv_703)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_705, %onnx::Conv_706)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_708, %onnx::Conv_709)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_711, %onnx::Conv_712)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_714, %onnx::Conv_715)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_717, %onnx::Conv_718)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_720, %onnx::Conv_721)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_723, %onnx::Conv_724)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/nl/Relu_output_0, %onnx::Conv_726, %onnx::Conv_727)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_729, %onnx::Conv_730)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_732, %onnx::Conv_733)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_735, %onnx::Conv_736)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_738, %onnx::Conv_739)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_741, %onnx::Conv_742)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 1104, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/nl/Relu_output_0, %onnx::Conv_744, %onnx::Conv_745)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_747, %onnx::Conv_748)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_750, %onnx::Conv_751)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%574 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %574
}
|
val_accuracy
| 0 | 84,016,000 | 2,672,316 |
{'zcp_synflow': 77.54000815330863, 'zcp_zen': 69.54060363769531, 'zcp_epe_nas': 33.66601130004994, 'zcp_fisher': 0.09900016337633133, 'zcp_flops': 84016000.0, 'zcp_grad_norm': 22.263874053955078, 'zcp_grasp': -0.022739410400390625, 'zcp_jacov': -16.057733628969903, 'zcp_l2_norm': 683.6270751953125, 'zcp_nwot': 212.26312003631605, 'zcp_params': 2672316.0, 'zcp_plain': 0.0012116212164983153, 'zcp_snip': 42.017982482910156, 'lat_1080ti_1': 0.5092561385754241, 'lat_1080ti_32': 0.4068873505690517, 'lat_1080ti_64': 0.335611908148044, 'lat_2080ti_1': 0.45078239366918876, 'lat_2080ti_32': 0.3490399758954854, 'lat_2080ti_64': 0.31837654298150775, 'lat_essential_ph_1': 0.33962264150943394, 'lat_eyeriss': 0.5930537472722935, 'lat_fpga': 0.6430204138176707, 'lat_gold_6226': 0.5908217443146769, 'lat_gold_6240': 0.6254641034799928, 'lat_pixel2': 0.41304347826086957, 'lat_pixel3': 0.5678333400259031, 'lat_raspi4': 0.6073549660401206, 'lat_samsung_a50': 0.2631578947368421, 'lat_samsung_s7': 0.23622047244094488, 'lat_silver_4114': 0.6860264414913013, 'lat_silver_4210r': 0.6054524578515099, 'lat_titan_rtx_1': 0.42626191463579594, 'lat_titan_rtx_32': 0.3574931308076952, 'lat_titan_rtx_64': 0.31798209234587976, 'lat_titanx_1': 0.2391031077432113, 'lat_titanx_32': 0.32161794708481517, 'lat_titanx_64': 0.33271968185051853, 'lat_titanxp_1': 0.4235740945051254, 'lat_titanxp_32': 0.335090458610136, 'lat_titanxp_64': 0.33041565741960416}
| |
FBNet_1619
|
FBNet
|
1619
|
1619
|
graph main_graph (
%input.1[FLOAT, 1x3x32x32]
%fc.weight[FLOAT, 100x1504]
%fc.bias[FLOAT, 100]
%onnx::Conv_676[FLOAT, 16x3x3x3]
%onnx::Conv_677[FLOAT, 16]
%onnx::Conv_679[FLOAT, 16x8x1x1]
%onnx::Conv_682[FLOAT, 16x1x3x3]
%onnx::Conv_685[FLOAT, 16x8x1x1]
%onnx::Conv_688[FLOAT, 16x8x1x1]
%onnx::Conv_691[FLOAT, 16x1x3x3]
%onnx::Conv_694[FLOAT, 24x8x1x1]
%onnx::Conv_695[FLOAT, 24]
%onnx::Conv_697[FLOAT, 72x24x1x1]
%onnx::Conv_698[FLOAT, 72]
%onnx::Conv_700[FLOAT, 72x1x3x3]
%onnx::Conv_703[FLOAT, 24x72x1x1]
%onnx::Conv_706[FLOAT, 144x24x1x1]
%onnx::Conv_707[FLOAT, 144]
%onnx::Conv_709[FLOAT, 144x1x3x3]
%onnx::Conv_712[FLOAT, 24x144x1x1]
%onnx::Conv_715[FLOAT, 24x12x1x1]
%onnx::Conv_718[FLOAT, 24x1x3x3]
%onnx::Conv_721[FLOAT, 24x12x1x1]
%onnx::Conv_724[FLOAT, 24x12x1x1]
%onnx::Conv_727[FLOAT, 24x1x3x3]
%onnx::Conv_730[FLOAT, 32x12x1x1]
%onnx::Conv_731[FLOAT, 32]
%onnx::Conv_733[FLOAT, 96x32x1x1]
%onnx::Conv_734[FLOAT, 96]
%onnx::Conv_736[FLOAT, 96x1x3x3]
%onnx::Conv_739[FLOAT, 32x96x1x1]
%onnx::Conv_742[FLOAT, 32x32x1x1]
%onnx::Conv_745[FLOAT, 32x1x3x3]
%onnx::Conv_748[FLOAT, 32x32x1x1]
%onnx::Conv_751[FLOAT, 192x32x1x1]
%onnx::Conv_752[FLOAT, 192]
%onnx::Conv_754[FLOAT, 192x1x3x3]
%onnx::Conv_757[FLOAT, 64x192x1x1]
%onnx::Conv_758[FLOAT, 64]
%onnx::Conv_760[FLOAT, 64x64x1x1]
%onnx::Conv_763[FLOAT, 64x1x5x5]
%onnx::Conv_766[FLOAT, 64x64x1x1]
%onnx::Conv_769[FLOAT, 64x64x1x1]
%onnx::Conv_772[FLOAT, 64x1x3x3]
%onnx::Conv_775[FLOAT, 64x64x1x1]
%onnx::Conv_778[FLOAT, 64x64x1x1]
%onnx::Conv_781[FLOAT, 64x1x5x5]
%onnx::Conv_784[FLOAT, 112x64x1x1]
%onnx::Conv_785[FLOAT, 112]
%onnx::Conv_787[FLOAT, 672x112x1x1]
%onnx::Conv_788[FLOAT, 672]
%onnx::Conv_790[FLOAT, 672x1x5x5]
%onnx::Conv_793[FLOAT, 112x672x1x1]
%onnx::Conv_796[FLOAT, 112x112x1x1]
%onnx::Conv_799[FLOAT, 112x1x5x5]
%onnx::Conv_802[FLOAT, 112x112x1x1]
%onnx::Conv_805[FLOAT, 672x112x1x1]
%onnx::Conv_808[FLOAT, 672x1x5x5]
%onnx::Conv_811[FLOAT, 112x672x1x1]
%onnx::Conv_814[FLOAT, 112x112x1x1]
%onnx::Conv_817[FLOAT, 112x1x3x3]
%onnx::Conv_820[FLOAT, 184x112x1x1]
%onnx::Conv_821[FLOAT, 184]
%onnx::Conv_823[FLOAT, 552x184x1x1]
%onnx::Conv_824[FLOAT, 552]
%onnx::Conv_826[FLOAT, 552x1x5x5]
%onnx::Conv_829[FLOAT, 184x552x1x1]
%onnx::Conv_832[FLOAT, 184x92x1x1]
%onnx::Conv_835[FLOAT, 184x1x5x5]
%onnx::Conv_838[FLOAT, 184x92x1x1]
%onnx::Conv_841[FLOAT, 552x184x1x1]
%onnx::Conv_844[FLOAT, 552x1x3x3]
%onnx::Conv_847[FLOAT, 184x552x1x1]
%onnx::Conv_850[FLOAT, 184x92x1x1]
%onnx::Conv_853[FLOAT, 184x1x3x3]
%onnx::Conv_856[FLOAT, 352x92x1x1]
%onnx::Conv_857[FLOAT, 352]
%onnx::Conv_859[FLOAT, 1504x352x1x1]
%onnx::Conv_860[FLOAT, 1504]
) {
%onnx::Conv_854 = Identity(%onnx::Conv_821)
%onnx::Conv_851 = Identity(%onnx::Conv_821)
%onnx::Conv_848 = Identity(%onnx::Conv_821)
%onnx::Conv_845 = Identity(%onnx::Conv_824)
%onnx::Conv_842 = Identity(%onnx::Conv_824)
%onnx::Conv_839 = Identity(%onnx::Conv_821)
%onnx::Conv_836 = Identity(%onnx::Conv_821)
%onnx::Conv_833 = Identity(%onnx::Conv_821)
%onnx::Conv_830 = Identity(%onnx::Conv_821)
%onnx::Conv_827 = Identity(%onnx::Conv_824)
%onnx::Conv_818 = Identity(%onnx::Conv_785)
%onnx::Conv_815 = Identity(%onnx::Conv_785)
%onnx::Conv_812 = Identity(%onnx::Conv_785)
%onnx::Conv_809 = Identity(%onnx::Conv_788)
%onnx::Conv_806 = Identity(%onnx::Conv_788)
%onnx::Conv_803 = Identity(%onnx::Conv_785)
%onnx::Conv_800 = Identity(%onnx::Conv_785)
%onnx::Conv_797 = Identity(%onnx::Conv_785)
%onnx::Conv_794 = Identity(%onnx::Conv_785)
%onnx::Conv_791 = Identity(%onnx::Conv_788)
%onnx::Conv_782 = Identity(%onnx::Conv_758)
%onnx::Conv_779 = Identity(%onnx::Conv_758)
%onnx::Conv_776 = Identity(%onnx::Conv_758)
%onnx::Conv_773 = Identity(%onnx::Conv_758)
%onnx::Conv_770 = Identity(%onnx::Conv_758)
%onnx::Conv_767 = Identity(%onnx::Conv_758)
%onnx::Conv_764 = Identity(%onnx::Conv_758)
%onnx::Conv_761 = Identity(%onnx::Conv_758)
%onnx::Conv_755 = Identity(%onnx::Conv_752)
%onnx::Conv_749 = Identity(%onnx::Conv_731)
%onnx::Conv_746 = Identity(%onnx::Conv_731)
%onnx::Conv_743 = Identity(%onnx::Conv_731)
%onnx::Conv_740 = Identity(%onnx::Conv_731)
%onnx::Conv_737 = Identity(%onnx::Conv_734)
%onnx::Conv_728 = Identity(%onnx::Conv_695)
%onnx::Conv_725 = Identity(%onnx::Conv_695)
%onnx::Conv_722 = Identity(%onnx::Conv_695)
%onnx::Conv_719 = Identity(%onnx::Conv_695)
%onnx::Conv_716 = Identity(%onnx::Conv_695)
%onnx::Conv_713 = Identity(%onnx::Conv_695)
%onnx::Conv_710 = Identity(%onnx::Conv_707)
%onnx::Conv_704 = Identity(%onnx::Conv_695)
%onnx::Conv_701 = Identity(%onnx::Conv_698)
%onnx::Conv_692 = Identity(%onnx::Conv_677)
%onnx::Conv_689 = Identity(%onnx::Conv_677)
%onnx::Conv_686 = Identity(%onnx::Conv_677)
%onnx::Conv_683 = Identity(%onnx::Conv_677)
%onnx::Conv_680 = Identity(%onnx::Conv_677)
%/stem/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%input.1, %onnx::Conv_676, %onnx::Conv_677)
%/stem/relu/Relu_output_0 = Relu(%/stem/conv/Conv_output_0)
%/cells.0/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/stem/relu/Relu_output_0, %onnx::Conv_679, %onnx::Conv_680)
%/cells.0/nl/Relu_output_0 = Relu(%/cells.0/conv1/Conv_output_0)
%/cells.0/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_output_0 = Reshape(%/cells.0/nl/Relu_output_0, %/cells.0/shuffle/Constant_output_0)
%/cells.0/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.0/shuffle/Reshape_output_0)
%/cells.0/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.0/shuffle/Reshape_1_output_0 = Reshape(%/cells.0/shuffle/Transpose_output_0, %/cells.0/shuffle/Constant_1_output_0)
%/cells.0/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.0/shuffle/Reshape_1_output_0, %onnx::Conv_682, %onnx::Conv_683)
%/cells.0/nl_1/Relu_output_0 = Relu(%/cells.0/conv2/Conv_output_0)
%/cells.0/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/nl_1/Relu_output_0, %onnx::Conv_685, %onnx::Conv_686)
%/cells.0/Add_output_0 = Add(%/cells.0/conv3/Conv_output_0, %/stem/relu/Relu_output_0)
%/cells.1/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.0/Add_output_0, %onnx::Conv_688, %onnx::Conv_689)
%/cells.1/nl/Relu_output_0 = Relu(%/cells.1/conv1/Conv_output_0)
%/cells.1/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_output_0 = Reshape(%/cells.1/nl/Relu_output_0, %/cells.1/shuffle/Constant_output_0)
%/cells.1/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.1/shuffle/Reshape_output_0)
%/cells.1/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.1/shuffle/Reshape_1_output_0 = Reshape(%/cells.1/shuffle/Transpose_output_0, %/cells.1/shuffle/Constant_1_output_0)
%/cells.1/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 16, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.1/shuffle/Reshape_1_output_0, %onnx::Conv_691, %onnx::Conv_692)
%/cells.1/nl_1/Relu_output_0 = Relu(%/cells.1/conv2/Conv_output_0)
%/cells.1/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/nl_1/Relu_output_0, %onnx::Conv_694, %onnx::Conv_695)
%/cells.2/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.1/conv3/Conv_output_0, %onnx::Conv_697, %onnx::Conv_698)
%/cells.2/nl/Relu_output_0 = Relu(%/cells.2/conv1/Conv_output_0)
%/cells.2/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 72, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.2/nl/Relu_output_0, %onnx::Conv_700, %onnx::Conv_701)
%/cells.2/nl_1/Relu_output_0 = Relu(%/cells.2/conv2/Conv_output_0)
%/cells.2/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/nl_1/Relu_output_0, %onnx::Conv_703, %onnx::Conv_704)
%/cells.2/Add_output_0 = Add(%/cells.2/conv3/Conv_output_0, %/cells.1/conv3/Conv_output_0)
%/cells.3/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.2/Add_output_0, %onnx::Conv_706, %onnx::Conv_707)
%/cells.3/nl/Relu_output_0 = Relu(%/cells.3/conv1/Conv_output_0)
%/cells.3/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 144, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.3/nl/Relu_output_0, %onnx::Conv_709, %onnx::Conv_710)
%/cells.3/nl_1/Relu_output_0 = Relu(%/cells.3/conv2/Conv_output_0)
%/cells.3/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/nl_1/Relu_output_0, %onnx::Conv_712, %onnx::Conv_713)
%/cells.3/Add_output_0 = Add(%/cells.3/conv3/Conv_output_0, %/cells.2/Add_output_0)
%/cells.4/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.3/Add_output_0, %onnx::Conv_715, %onnx::Conv_716)
%/cells.4/nl/Relu_output_0 = Relu(%/cells.4/conv1/Conv_output_0)
%/cells.4/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_output_0 = Reshape(%/cells.4/nl/Relu_output_0, %/cells.4/shuffle/Constant_output_0)
%/cells.4/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.4/shuffle/Reshape_output_0)
%/cells.4/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.4/shuffle/Reshape_1_output_0 = Reshape(%/cells.4/shuffle/Transpose_output_0, %/cells.4/shuffle/Constant_1_output_0)
%/cells.4/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.4/shuffle/Reshape_1_output_0, %onnx::Conv_718, %onnx::Conv_719)
%/cells.4/nl_1/Relu_output_0 = Relu(%/cells.4/conv2/Conv_output_0)
%/cells.4/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/nl_1/Relu_output_0, %onnx::Conv_721, %onnx::Conv_722)
%/cells.4/Add_output_0 = Add(%/cells.4/conv3/Conv_output_0, %/cells.3/Add_output_0)
%/cells.5/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.4/Add_output_0, %onnx::Conv_724, %onnx::Conv_725)
%/cells.5/nl/Relu_output_0 = Relu(%/cells.5/conv1/Conv_output_0)
%/cells.5/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_output_0 = Reshape(%/cells.5/nl/Relu_output_0, %/cells.5/shuffle/Constant_output_0)
%/cells.5/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.5/shuffle/Reshape_output_0)
%/cells.5/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.5/shuffle/Reshape_1_output_0 = Reshape(%/cells.5/shuffle/Transpose_output_0, %/cells.5/shuffle/Constant_1_output_0)
%/cells.5/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 24, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.5/shuffle/Reshape_1_output_0, %onnx::Conv_727, %onnx::Conv_728)
%/cells.5/nl_1/Relu_output_0 = Relu(%/cells.5/conv2/Conv_output_0)
%/cells.5/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/nl_1/Relu_output_0, %onnx::Conv_730, %onnx::Conv_731)
%/cells.7/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.5/conv3/Conv_output_0, %onnx::Conv_733, %onnx::Conv_734)
%/cells.7/nl/Relu_output_0 = Relu(%/cells.7/conv1/Conv_output_0)
%/cells.7/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.7/nl/Relu_output_0, %onnx::Conv_736, %onnx::Conv_737)
%/cells.7/nl_1/Relu_output_0 = Relu(%/cells.7/conv2/Conv_output_0)
%/cells.7/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/nl_1/Relu_output_0, %onnx::Conv_739, %onnx::Conv_740)
%/cells.7/Add_output_0 = Add(%/cells.7/conv3/Conv_output_0, %/cells.5/conv3/Conv_output_0)
%/cells.8/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.7/Add_output_0, %onnx::Conv_742, %onnx::Conv_743)
%/cells.8/nl/Relu_output_0 = Relu(%/cells.8/conv1/Conv_output_0)
%/cells.8/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 32, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.8/nl/Relu_output_0, %onnx::Conv_745, %onnx::Conv_746)
%/cells.8/nl_1/Relu_output_0 = Relu(%/cells.8/conv2/Conv_output_0)
%/cells.8/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/nl_1/Relu_output_0, %onnx::Conv_748, %onnx::Conv_749)
%/cells.8/Add_output_0 = Add(%/cells.8/conv3/Conv_output_0, %/cells.7/Add_output_0)
%/cells.9/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.8/Add_output_0, %onnx::Conv_751, %onnx::Conv_752)
%/cells.9/nl/Relu_output_0 = Relu(%/cells.9/conv1/Conv_output_0)
%/cells.9/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 192, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.9/nl/Relu_output_0, %onnx::Conv_754, %onnx::Conv_755)
%/cells.9/nl_1/Relu_output_0 = Relu(%/cells.9/conv2/Conv_output_0)
%/cells.9/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/nl_1/Relu_output_0, %onnx::Conv_757, %onnx::Conv_758)
%/cells.10/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.9/conv3/Conv_output_0, %onnx::Conv_760, %onnx::Conv_761)
%/cells.10/nl/Relu_output_0 = Relu(%/cells.10/conv1/Conv_output_0)
%/cells.10/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.10/nl/Relu_output_0, %onnx::Conv_763, %onnx::Conv_764)
%/cells.10/nl_1/Relu_output_0 = Relu(%/cells.10/conv2/Conv_output_0)
%/cells.10/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/nl_1/Relu_output_0, %onnx::Conv_766, %onnx::Conv_767)
%/cells.10/Add_output_0 = Add(%/cells.10/conv3/Conv_output_0, %/cells.9/conv3/Conv_output_0)
%/cells.12/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.10/Add_output_0, %onnx::Conv_769, %onnx::Conv_770)
%/cells.12/nl/Relu_output_0 = Relu(%/cells.12/conv1/Conv_output_0)
%/cells.12/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.12/nl/Relu_output_0, %onnx::Conv_772, %onnx::Conv_773)
%/cells.12/nl_1/Relu_output_0 = Relu(%/cells.12/conv2/Conv_output_0)
%/cells.12/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/nl_1/Relu_output_0, %onnx::Conv_775, %onnx::Conv_776)
%/cells.12/Add_output_0 = Add(%/cells.12/conv3/Conv_output_0, %/cells.10/Add_output_0)
%/cells.13/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.12/Add_output_0, %onnx::Conv_778, %onnx::Conv_779)
%/cells.13/nl/Relu_output_0 = Relu(%/cells.13/conv1/Conv_output_0)
%/cells.13/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 64, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.13/nl/Relu_output_0, %onnx::Conv_781, %onnx::Conv_782)
%/cells.13/nl_1/Relu_output_0 = Relu(%/cells.13/conv2/Conv_output_0)
%/cells.13/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/nl_1/Relu_output_0, %onnx::Conv_784, %onnx::Conv_785)
%/cells.14/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.13/conv3/Conv_output_0, %onnx::Conv_787, %onnx::Conv_788)
%/cells.14/nl/Relu_output_0 = Relu(%/cells.14/conv1/Conv_output_0)
%/cells.14/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.14/nl/Relu_output_0, %onnx::Conv_790, %onnx::Conv_791)
%/cells.14/nl_1/Relu_output_0 = Relu(%/cells.14/conv2/Conv_output_0)
%/cells.14/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/nl_1/Relu_output_0, %onnx::Conv_793, %onnx::Conv_794)
%/cells.14/Add_output_0 = Add(%/cells.14/conv3/Conv_output_0, %/cells.13/conv3/Conv_output_0)
%/cells.15/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.14/Add_output_0, %onnx::Conv_796, %onnx::Conv_797)
%/cells.15/nl/Relu_output_0 = Relu(%/cells.15/conv1/Conv_output_0)
%/cells.15/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.15/nl/Relu_output_0, %onnx::Conv_799, %onnx::Conv_800)
%/cells.15/nl_1/Relu_output_0 = Relu(%/cells.15/conv2/Conv_output_0)
%/cells.15/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/nl_1/Relu_output_0, %onnx::Conv_802, %onnx::Conv_803)
%/cells.15/Add_output_0 = Add(%/cells.15/conv3/Conv_output_0, %/cells.14/Add_output_0)
%/cells.16/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.15/Add_output_0, %onnx::Conv_805, %onnx::Conv_806)
%/cells.16/nl/Relu_output_0 = Relu(%/cells.16/conv1/Conv_output_0)
%/cells.16/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 672, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.16/nl/Relu_output_0, %onnx::Conv_808, %onnx::Conv_809)
%/cells.16/nl_1/Relu_output_0 = Relu(%/cells.16/conv2/Conv_output_0)
%/cells.16/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/nl_1/Relu_output_0, %onnx::Conv_811, %onnx::Conv_812)
%/cells.16/Add_output_0 = Add(%/cells.16/conv3/Conv_output_0, %/cells.15/Add_output_0)
%/cells.17/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.16/Add_output_0, %onnx::Conv_814, %onnx::Conv_815)
%/cells.17/nl/Relu_output_0 = Relu(%/cells.17/conv1/Conv_output_0)
%/cells.17/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 112, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%/cells.17/nl/Relu_output_0, %onnx::Conv_817, %onnx::Conv_818)
%/cells.17/nl_1/Relu_output_0 = Relu(%/cells.17/conv2/Conv_output_0)
%/cells.17/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/nl_1/Relu_output_0, %onnx::Conv_820, %onnx::Conv_821)
%/cells.18/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.17/conv3/Conv_output_0, %onnx::Conv_823, %onnx::Conv_824)
%/cells.18/nl/Relu_output_0 = Relu(%/cells.18/conv1/Conv_output_0)
%/cells.18/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.18/nl/Relu_output_0, %onnx::Conv_826, %onnx::Conv_827)
%/cells.18/nl_1/Relu_output_0 = Relu(%/cells.18/conv2/Conv_output_0)
%/cells.18/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/nl_1/Relu_output_0, %onnx::Conv_829, %onnx::Conv_830)
%/cells.18/Add_output_0 = Add(%/cells.18/conv3/Conv_output_0, %/cells.17/conv3/Conv_output_0)
%/cells.19/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.18/Add_output_0, %onnx::Conv_832, %onnx::Conv_833)
%/cells.19/nl/Relu_output_0 = Relu(%/cells.19/conv1/Conv_output_0)
%/cells.19/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_output_0 = Reshape(%/cells.19/nl/Relu_output_0, %/cells.19/shuffle/Constant_output_0)
%/cells.19/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.19/shuffle/Reshape_output_0)
%/cells.19/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.19/shuffle/Reshape_1_output_0 = Reshape(%/cells.19/shuffle/Transpose_output_0, %/cells.19/shuffle/Constant_1_output_0)
%/cells.19/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [5, 5], pads = [2, 2, 2, 2], strides = [1, 1]](%/cells.19/shuffle/Reshape_1_output_0, %onnx::Conv_835, %onnx::Conv_836)
%/cells.19/nl_1/Relu_output_0 = Relu(%/cells.19/conv2/Conv_output_0)
%/cells.19/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/nl_1/Relu_output_0, %onnx::Conv_838, %onnx::Conv_839)
%/cells.19/Add_output_0 = Add(%/cells.19/conv3/Conv_output_0, %/cells.18/Add_output_0)
%/cells.20/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.19/Add_output_0, %onnx::Conv_841, %onnx::Conv_842)
%/cells.20/nl/Relu_output_0 = Relu(%/cells.20/conv1/Conv_output_0)
%/cells.20/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 552, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.20/nl/Relu_output_0, %onnx::Conv_844, %onnx::Conv_845)
%/cells.20/nl_1/Relu_output_0 = Relu(%/cells.20/conv2/Conv_output_0)
%/cells.20/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/nl_1/Relu_output_0, %onnx::Conv_847, %onnx::Conv_848)
%/cells.20/Add_output_0 = Add(%/cells.20/conv3/Conv_output_0, %/cells.19/Add_output_0)
%/cells.21/conv1/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.20/Add_output_0, %onnx::Conv_850, %onnx::Conv_851)
%/cells.21/nl/Relu_output_0 = Relu(%/cells.21/conv1/Conv_output_0)
%/cells.21/shuffle/Constant_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_output_0 = Reshape(%/cells.21/nl/Relu_output_0, %/cells.21/shuffle/Constant_output_0)
%/cells.21/shuffle/Transpose_output_0 = Transpose[perm = [0, 2, 1, 3, 4]](%/cells.21/shuffle/Reshape_output_0)
%/cells.21/shuffle/Constant_1_output_0 = Constant[value = <Tensor>]()
%/cells.21/shuffle/Reshape_1_output_0 = Reshape(%/cells.21/shuffle/Transpose_output_0, %/cells.21/shuffle/Constant_1_output_0)
%/cells.21/conv2/Conv_output_0 = Conv[dilations = [1, 1], group = 184, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%/cells.21/shuffle/Reshape_1_output_0, %onnx::Conv_853, %onnx::Conv_854)
%/cells.21/nl_1/Relu_output_0 = Relu(%/cells.21/conv2/Conv_output_0)
%/cells.21/conv3/Conv_output_0 = Conv[dilations = [1, 1], group = 2, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/nl_1/Relu_output_0, %onnx::Conv_856, %onnx::Conv_857)
%/header/conv/Conv_output_0 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%/cells.21/conv3/Conv_output_0, %onnx::Conv_859, %onnx::Conv_860)
%/header/relu/Relu_output_0 = Relu(%/header/conv/Conv_output_0)
%/avgpool/GlobalAveragePool_output_0 = GlobalAveragePool(%/header/relu/Relu_output_0)
%/Constant_output_0 = Constant[value = <Tensor>]()
%/Reshape_output_0 = Reshape(%/avgpool/GlobalAveragePool_output_0, %/Constant_output_0)
%674 = Gemm[alpha = 1, beta = 1, transB = 1](%/Reshape_output_0, %fc.weight, %fc.bias)
return %674
}
|
val_accuracy
| 0 | 66,235,776 | 1,689,596 |
{'zcp_synflow': 73.14632019089242, 'zcp_zen': 65.00640869140625, 'zcp_epe_nas': 5.849681685330725, 'zcp_fisher': 0.09445938467979431, 'zcp_flops': 66235776.0, 'zcp_grad_norm': 18.732200622558594, 'zcp_grasp': -0.07204627990722656, 'zcp_jacov': -16.06270800667419, 'zcp_l2_norm': 591.088623046875, 'zcp_nwot': 211.61124226532692, 'zcp_params': 1689596.0, 'zcp_plain': 0.005323455668985844, 'zcp_snip': 35.63189697265625, 'lat_1080ti_1': 0.5319214748632429, 'lat_1080ti_32': 0.5148264303480484, 'lat_1080ti_64': 0.4349706381599573, 'lat_2080ti_1': 0.6301407055699206, 'lat_2080ti_32': 0.5747147968051503, 'lat_2080ti_64': 0.4757870380864777, 'lat_essential_ph_1': 0.24528301886792453, 'lat_eyeriss': 0.37149065703335576, 'lat_fpga': 0.4666976716571507, 'lat_gold_6226': 0.29621071084133255, 'lat_gold_6240': 0.45888345860688695, 'lat_pixel2': 0.2826086956521739, 'lat_pixel3': 0.3983101217814515, 'lat_raspi4': 0.3872696646437584, 'lat_samsung_a50': 0.16842105263157894, 'lat_samsung_s7': 0.2204724409448819, 'lat_silver_4114': 0.5074924263199182, 'lat_silver_4210r': 0.5004321004994156, 'lat_titan_rtx_1': 0.5980211129027823, 'lat_titan_rtx_32': 0.5642644826102936, 'lat_titan_rtx_64': 0.5034510452666765, 'lat_titanx_1': 0.31995086618801727, 'lat_titanx_32': 0.5026608229195249, 'lat_titanx_64': 0.4682647224841943, 'lat_titanxp_1': 0.5756554674970784, 'lat_titanxp_32': 0.531354416793016, 'lat_titanxp_64': 0.46163783360661503}
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.