Set triton kernels as default. Remove deprecated config fields. (#5)
Browse files- Set triton kernels as default. Remove deprecated config fields. (81d49bfd671aa2304153deed6d34771d424de183)
Co-authored-by: Maximilian Beck <[email protected]>
- config.json +12 -5
config.json
CHANGED
|
@@ -1,5 +1,5 @@
|
|
| 1 |
{
|
| 2 |
-
"_name_or_path": "
|
| 3 |
"add_embedding_dropout": false,
|
| 4 |
"add_forward_backend_padding": false,
|
| 5 |
"add_out_norm": true,
|
|
@@ -9,18 +9,23 @@
|
|
| 9 |
"architectures": [
|
| 10 |
"xLSTMForCausalLM"
|
| 11 |
],
|
|
|
|
| 12 |
"bos_token_id": 0,
|
| 13 |
"cell_norm_eps": 1e-06,
|
|
|
|
|
|
|
| 14 |
"embedding_dim": 4096,
|
| 15 |
"eos_token_id": 2,
|
|
|
|
| 16 |
"ffn_proj_factor": 2.667,
|
| 17 |
"ffn_round_up_to_multiple_of": 64,
|
| 18 |
"force_bos_token_insert": true,
|
| 19 |
-
"forward_backend_name": "chunkwise--triton_limit_chunk",
|
| 20 |
"gate_soft_cap": 15.0,
|
| 21 |
"head_dim": 512,
|
| 22 |
"igate_bias_init_range": -10.0,
|
|
|
|
| 23 |
"mlstm_round_up_to_multiple_of": 64,
|
|
|
|
| 24 |
"model_type": "xlstm",
|
| 25 |
"norm_eps": 1e-06,
|
| 26 |
"norm_reduction_force_float32": true,
|
|
@@ -30,12 +35,14 @@
|
|
| 30 |
"pad_token_id": 1,
|
| 31 |
"qk_dim_factor": 0.5,
|
| 32 |
"return_last_states": true,
|
| 33 |
-
"
|
|
|
|
| 34 |
"tie_word_embeddings": false,
|
| 35 |
"torch_dtype": "float32",
|
| 36 |
-
"transformers_version": "4.
|
| 37 |
"use_bias": false,
|
| 38 |
"use_cache": true,
|
| 39 |
"v_dim_factor": 1.0,
|
| 40 |
-
"vocab_size": 50304
|
|
|
|
| 41 |
}
|
|
|
|
| 1 |
{
|
| 2 |
+
"_name_or_path": "NX-AI/xLSTM-7b",
|
| 3 |
"add_embedding_dropout": false,
|
| 4 |
"add_forward_backend_padding": false,
|
| 5 |
"add_out_norm": true,
|
|
|
|
| 9 |
"architectures": [
|
| 10 |
"xLSTMForCausalLM"
|
| 11 |
],
|
| 12 |
+
"autocast_kernel_dtype": "bfloat16",
|
| 13 |
"bos_token_id": 0,
|
| 14 |
"cell_norm_eps": 1e-06,
|
| 15 |
+
"chunk_size": 64,
|
| 16 |
+
"chunkwise_kernel": "chunkwise--triton_xl_chunk",
|
| 17 |
"embedding_dim": 4096,
|
| 18 |
"eos_token_id": 2,
|
| 19 |
+
"eps": 1e-06,
|
| 20 |
"ffn_proj_factor": 2.667,
|
| 21 |
"ffn_round_up_to_multiple_of": 64,
|
| 22 |
"force_bos_token_insert": true,
|
|
|
|
| 23 |
"gate_soft_cap": 15.0,
|
| 24 |
"head_dim": 512,
|
| 25 |
"igate_bias_init_range": -10.0,
|
| 26 |
+
"inference_state_dtype": "float32",
|
| 27 |
"mlstm_round_up_to_multiple_of": 64,
|
| 28 |
+
"mode": "inference",
|
| 29 |
"model_type": "xlstm",
|
| 30 |
"norm_eps": 1e-06,
|
| 31 |
"norm_reduction_force_float32": true,
|
|
|
|
| 35 |
"pad_token_id": 1,
|
| 36 |
"qk_dim_factor": 0.5,
|
| 37 |
"return_last_states": true,
|
| 38 |
+
"sequence_kernel": "native_sequence__triton",
|
| 39 |
+
"step_kernel": "triton",
|
| 40 |
"tie_word_embeddings": false,
|
| 41 |
"torch_dtype": "float32",
|
| 42 |
+
"transformers_version": "4.48.0.dev0",
|
| 43 |
"use_bias": false,
|
| 44 |
"use_cache": true,
|
| 45 |
"v_dim_factor": 1.0,
|
| 46 |
+
"vocab_size": 50304,
|
| 47 |
+
"weight_mode": "single"
|
| 48 |
}
|