Upload folder using huggingface_hub

config.json

@@ -1,5 +1,4 @@
 {
-  "_name_or_path": "togethercomputer/RedPajama-INCITE-Instruct-3B-v1",
   "architectures": [
     "GPTNeoXForCausalLM"
   ],
@@ -18,11 +17,14 @@
   "model_type": "gpt_neox",
   "num_attention_heads": 32,
   "num_hidden_layers": 32,
+  "partial_rotary_factor": 1.0,
   "rope_scaling": null,
+  "rope_theta": 10000,
   "rotary_emb_base": 10000,
   "rotary_pct": 1.0,
   "tie_word_embeddings": false,
-  "transformers_version": "4.41.2",
+  "torch_dtype": "float16",
+  "transformers_version": "4.52.4",
   "use_cache": true,
   "use_parallel_residual": false,
   "vocab_size": 50432

generation_config.json

@@ -2,5 +2,5 @@
   "_from_model_config": true,
   "bos_token_id": 0,
   "eos_token_id": 0,
-  "transformers_version": "4.41.2"
+  "transformers_version": "4.52.4"
 }

openvino_detokenizer.xml

@@ -1,16 +1,16 @@
 <?xml version="1.0"?>
 <net name="detokenizer" version="11">
 	<layers>
-		<layer id="0" name="Parameter_133724" type="Parameter" version="opset1">
+		<layer id="0" name="Parameter_135008" type="Parameter" version="opset1">
 			<data shape="?,?" element_type="i64" />
 			<output>
-				<port id="0" precision="I64" names="Parameter_133724">
+				<port id="0" precision="I64" names="Parameter_135008">
 					<dim>-1</dim>
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1" name="Convert_133919" type="Convert" version="opset1">
+		<layer id="1" name="Convert_135203" type="Convert" version="opset1">
 			<data destination_type="i32" />
 			<input>
 				<port id="0" precision="I64">
@@ -25,7 +25,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="2" name="Constant_133726" type="Const" version="opset1">
+		<layer id="2" name="Constant_135010" type="Const" version="opset1">
 			<data element_type="i32" shape="50277" offset="0" size="201108" />
 			<output>
 				<port id="0" precision="I32">
@@ -33,7 +33,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="3" name="Constant_133728" type="Const" version="opset1">
+		<layer id="3" name="Constant_135012" type="Const" version="opset1">
 			<data element_type="i32" shape="50277" offset="201108" size="201108" />
 			<output>
 				<port id="0" precision="I32">
@@ -41,7 +41,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="4" name="Constant_133730" type="Const" version="opset1">
+		<layer id="4" name="Constant_135014" type="Const" version="opset1">
 			<data element_type="u8" shape="312914" offset="402216" size="312914" />
 			<output>
 				<port id="0" precision="U8">
@@ -49,7 +49,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="5" name="Slice_133735" type="Const" version="opset1">
+		<layer id="5" name="Slice_135019" type="Const" version="opset1">
 			<data element_type="i32" shape="2" offset="715130" size="8" />
 			<output>
 				<port id="0" precision="I32">
@@ -57,7 +57,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="6" name="VocabDecoder_133737" type="VocabDecoder" version="extension">
+		<layer id="6" name="VocabDecoder_135021" type="VocabDecoder" version="extension">
 			<data skip_tokens="" />
 			<input>
 				<port id="0" precision="I32">
@@ -95,7 +95,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="7" name="FuzeRagged_133738" type="FuzeRagged" version="extension">
+		<layer id="7" name="FuzeRagged_135022" type="FuzeRagged" version="extension">
 			<input>
 				<port id="0" precision="I32">
 					<dim>-1</dim>
@@ -119,7 +119,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="8" name="UTF8Validate_133739" type="UTF8Validate" version="extension">
+		<layer id="8" name="UTF8Validate_135023" type="UTF8Validate" version="extension">
 			<data replace_mode="true" />
 			<input>
 				<port id="0" precision="I32">
@@ -144,7 +144,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="9" name="Constant_133741" type="Const" version="opset1">
+		<layer id="9" name="Constant_135025" type="Const" version="opset1">
 			<data element_type="u8" shape="51" offset="715138" size="51" />
 			<output>
 				<port id="0" precision="U8">
@@ -152,7 +152,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="10" name="Constant_133743" type="Const" version="opset1">
+		<layer id="10" name="Constant_135027" type="Const" version="opset1">
 			<data element_type="u8" shape="2" offset="715189" size="2" />
 			<output>
 				<port id="0" precision="U8">
@@ -160,7 +160,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="11" name="RegexNormalization_133744" type="RegexNormalization" version="extension">
+		<layer id="11" name="RegexNormalization_135028" type="RegexNormalization" version="extension">
 			<data global_replace="true" />
 			<input>
 				<port id="0" precision="I32">
@@ -191,7 +191,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="12" name="StringTensorPack_133745" type="StringTensorPack" version="opset15">
+		<layer id="12" name="StringTensorPack_135029" type="StringTensorPack" version="opset15">
 			<input>
 				<port id="0" precision="I32">
 					<dim>-1</dim>
@@ -204,12 +204,12 @@
 				</port>
 			</input>
 			<output>
-				<port id="3" precision="STRING" names="Result_133746,string_output">
+				<port id="3" precision="STRING" names="Result_135030,string_output">
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="13" name="Result_133746" type="Result" version="opset1" output_names="Result_133746,string_output">
+		<layer id="13" name="Result_135030" type="Result" version="opset1" output_names="Result_135030,string_output">
 			<input>
 				<port id="0" precision="STRING">
 					<dim>-1</dim>

openvino_model.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e17c6e844a9f1ff1b4a65a7a079a82d09ca2f0db2b6b60ca54197d98bdcdc51d
-size 5556578690
+oid sha256:d89f150b2654a56bd302a94bbc0e6da14c45cecebc1c72a9b552b12146996604
+size 5551728902

openvino_tokenizer.xml

@@ -1,21 +1,21 @@
 <?xml version="1.0"?>
 <net name="tokenizer" version="11">
 	<layers>
-		<layer id="0" name="Parameter_133598" type="Parameter" version="opset1">
+		<layer id="0" name="Parameter_134882" type="Parameter" version="opset1">
 			<data shape="?" element_type="string" />
 			<output>
-				<port id="0" precision="STRING" names="Parameter_133598">
+				<port id="0" precision="STRING" names="Parameter_134882">
 					<dim>-1</dim>
 				</port>
 			</output>
 		</layer>
-		<layer id="1" name="Constant_133604" type="Const" version="opset1">
+		<layer id="1" name="Constant_134888" type="Const" version="opset1">
 			<data element_type="i64" shape="" offset="0" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="2" name="StringTensorUnpack_133599" type="StringTensorUnpack" version="opset15">
+		<layer id="2" name="StringTensorUnpack_134883" type="StringTensorUnpack" version="opset15">
 			<input>
 				<port id="0" precision="STRING">
 					<dim>-1</dim>
@@ -33,7 +33,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="3" name="ShapeOf_133600" type="ShapeOf" version="opset3">
+		<layer id="3" name="ShapeOf_134884" type="ShapeOf" version="opset3">
 			<data output_type="i64" />
 			<input>
 				<port id="0" precision="I32">
@@ -46,19 +46,19 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="4" name="Constant_133601" type="Const" version="opset1">
+		<layer id="4" name="Constant_134885" type="Const" version="opset1">
 			<data element_type="i64" shape="" offset="0" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="5" name="Constant_133602" type="Const" version="opset1">
+		<layer id="5" name="Constant_134886" type="Const" version="opset1">
 			<data element_type="i64" shape="" offset="0" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="6" name="Gather_133603" type="Gather" version="opset8">
+		<layer id="6" name="Gather_134887" type="Gather" version="opset8">
 			<data batch_dims="0" />
 			<input>
 				<port id="0" precision="I64">
@@ -71,13 +71,13 @@
 				<port id="3" precision="I64" />
 			</output>
 		</layer>
-		<layer id="7" name="Constant_133605" type="Const" version="opset1">
+		<layer id="7" name="Constant_134889" type="Const" version="opset1">
 			<data element_type="i64" shape="" offset="8" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="8" name="Range_133606" type="Range" version="opset4">
+		<layer id="8" name="Range_134890" type="Range" version="opset4">
 			<data output_type="i32" />
 			<input>
 				<port id="0" precision="I64" />
@@ -90,19 +90,19 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="9" name="Constant_133607" type="Const" version="opset1">
+		<layer id="9" name="Constant_134891" type="Const" version="opset1">
 			<data element_type="i64" shape="" offset="8" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="10" name="Constant_133608" type="Const" version="opset1">
+		<layer id="10" name="Constant_134892" type="Const" version="opset1">
 			<data element_type="i64" shape="" offset="8" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="11" name="Add_133609" type="Add" version="opset1">
+		<layer id="11" name="Add_134893" type="Add" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="I64" />
@@ -112,13 +112,13 @@
 				<port id="2" precision="I64" />
 			</output>
 		</layer>
-		<layer id="12" name="Constant_133610" type="Const" version="opset1">
+		<layer id="12" name="Constant_134894" type="Const" version="opset1">
 			<data element_type="i64" shape="" offset="8" size="8" />
 			<output>
 				<port id="0" precision="I64" />
 			</output>
 		</layer>
-		<layer id="13" name="Range_133611" type="Range" version="opset4">
+		<layer id="13" name="Range_134895" type="Range" version="opset4">
 			<data output_type="i32" />
 			<input>
 				<port id="0" precision="I64" />
@@ -131,7 +131,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="14" name="Constant_133673" type="Const" version="opset1">
+		<layer id="14" name="Constant_134957" type="Const" version="opset1">
 			<data element_type="u8" shape="704" offset="16" size="704" />
 			<output>
 				<port id="0" precision="U8">
@@ -139,7 +139,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="15" name="SpecialTokensSplit_133674" type="SpecialTokensSplit" version="extension">
+		<layer id="15" name="SpecialTokensSplit_134958" type="SpecialTokensSplit" version="extension">
 			<input>
 				<port id="0" precision="I32">
 					<dim>-1</dim>
@@ -181,7 +181,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="16" name="CharsMapNormalization_133675" type="CharsMapNormalization" version="extension">
+		<layer id="16" name="CharsMapNormalization_134959" type="CharsMapNormalization" version="extension">
 			<data add_dummy_prefix="false" remove_extra_whitespaces="false" escape_whitespaces="false" normalization_form="nfc" case_fold="false" nmt="false" />
 			<input>
 				<port id="0" precision="I32">
@@ -212,7 +212,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="17" name="Constant_133677" type="Const" version="opset1">
+		<layer id="17" name="Constant_134961" type="Const" version="opset1">
 			<data element_type="u8" shape="74" offset="720" size="74" />
 			<output>
 				<port id="0" precision="U8">
@@ -220,7 +220,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="18" name="RegexSplit_133678" type="RegexSplit" version="extension">
+		<layer id="18" name="RegexSplit_134962" type="RegexSplit" version="extension">
 			<data behaviour="isolate" invert="false" max_splits="-1" />
 			<input>
 				<port id="0" precision="I32">
@@ -266,7 +266,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="19" name="Constant_133680" type="Const" version="opset1">
+		<layer id="19" name="Constant_134964" type="Const" version="opset1">
 			<data element_type="i32" shape="50277" offset="794" size="201108" />
 			<output>
 				<port id="0" precision="I32">
@@ -274,7 +274,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="20" name="Constant_133682" type="Const" version="opset1">
+		<layer id="20" name="Constant_134966" type="Const" version="opset1">
 			<data element_type="i32" shape="50277" offset="201902" size="201108" />
 			<output>
 				<port id="0" precision="I32">
@@ -282,7 +282,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="21" name="Constant_133684" type="Const" version="opset1">
+		<layer id="21" name="Constant_134968" type="Const" version="opset1">
 			<data element_type="u8" shape="312914" offset="403010" size="312914" />
 			<output>
 				<port id="0" precision="U8">
@@ -290,7 +290,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="22" name="Constant_133692" type="Const" version="opset1">
+		<layer id="22" name="Constant_134976" type="Const" version="opset1">
 			<data element_type="i32" shape="50009" offset="715924" size="200036" />
 			<output>
 				<port id="0" precision="I32">
@@ -298,7 +298,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="23" name="Constant_133694" type="Const" version="opset1">
+		<layer id="23" name="Constant_134978" type="Const" version="opset1">
 			<data element_type="i32" shape="50009" offset="915960" size="200036" />
 			<output>
 				<port id="0" precision="I32">
@@ -306,7 +306,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="24" name="Constant_133696" type="Const" version="opset1">
+		<layer id="24" name="Constant_134980" type="Const" version="opset1">
 			<data element_type="u8" shape="162892" offset="1115996" size="162892" />
 			<output>
 				<port id="0" precision="U8">
@@ -314,7 +314,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="25" name="Constant_133698" type="Const" version="opset1">
+		<layer id="25" name="Constant_134982" type="Const" version="opset1">
 			<data element_type="i32" shape="50009" offset="1278888" size="200036" />
 			<output>
 				<port id="0" precision="I32">
@@ -322,7 +322,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="26" name="Constant_133700" type="Const" version="opset1">
+		<layer id="26" name="Constant_134984" type="Const" version="opset1">
 			<data element_type="i32" shape="50009" offset="1478924" size="200036" />
 			<output>
 				<port id="0" precision="I32">
@@ -330,7 +330,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="27" name="Constant_133702" type="Const" version="opset1">
+		<layer id="27" name="Constant_134986" type="Const" version="opset1">
 			<data element_type="u8" shape="149456" offset="1678960" size="149456" />
 			<output>
 				<port id="0" precision="U8">
@@ -338,7 +338,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="28" name="Constant_133686" type="Const" version="opset1">
+		<layer id="28" name="Constant_134970" type="Const" version="opset1">
 			<data element_type="i32" shape="24" offset="1828416" size="96" />
 			<output>
 				<port id="0" precision="I32">
@@ -346,7 +346,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="29" name="Constant_133688" type="Const" version="opset1">
+		<layer id="29" name="Constant_134972" type="Const" version="opset1">
 			<data element_type="i32" shape="24" offset="1828512" size="96" />
 			<output>
 				<port id="0" precision="I32">
@@ -354,7 +354,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="30" name="Constant_133690" type="Const" version="opset1">
+		<layer id="30" name="Constant_134974" type="Const" version="opset1">
 			<data element_type="u8" shape="310" offset="1828608" size="310" />
 			<output>
 				<port id="0" precision="U8">
@@ -362,7 +362,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="31" name="Constant_133703" type="Const" version="opset1">
+		<layer id="31" name="Constant_134987" type="Const" version="opset1">
 			<data element_type="i32" shape="24" offset="1828918" size="96" />
 			<output>
 				<port id="0" precision="I32">
@@ -370,7 +370,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="32" name="BPETokenizer_133704" type="BPETokenizer" version="extension">
+		<layer id="32" name="BPETokenizer_134988" type="BPETokenizer" version="extension">
 			<data unk_token="" fuse_unk="false" suffix_indicator="" end_suffix="" byte_fallback="false" cache_capacity="20000" />
 			<input>
 				<port id="0" precision="I32">
@@ -440,13 +440,13 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="33" name="Constant_133705" type="Const" version="opset1">
+		<layer id="33" name="Constant_134989" type="Const" version="opset1">
 			<data element_type="i32" shape="" offset="1829014" size="4" />
 			<output>
 				<port id="0" precision="I32" />
 			</output>
 		</layer>
-		<layer id="34" name="Constant_133707" type="Const" version="opset1">
+		<layer id="34" name="Constant_134991" type="Const" version="opset1">
 			<data element_type="u8" shape="4" offset="1829018" size="4" />
 			<output>
 				<port id="0" precision="U8">
@@ -454,7 +454,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="35" name="Constant_133709" type="Const" version="opset1">
+		<layer id="35" name="Constant_134993" type="Const" version="opset1">
 			<data element_type="u8" shape="13" offset="1829022" size="13" />
 			<output>
 				<port id="0" precision="U8">
@@ -462,7 +462,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="36" name="Truncate_133710" type="Truncate" version="extension">
+		<layer id="36" name="Truncate_134994" type="Truncate" version="extension">
 			<data m_num_inputs="1" />
 			<input>
 				<port id="0" precision="I32">
@@ -494,7 +494,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="37" name="Constant_133711" type="Const" version="opset1">
+		<layer id="37" name="Constant_134995" type="Const" version="opset1">
 			<data element_type="i32" shape="1" offset="1829035" size="4" />
 			<output>
 				<port id="0" precision="I32">
@@ -502,7 +502,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="38" name="CombineSegments_133712" type="CombineSegments" version="extension">
+		<layer id="38" name="CombineSegments_134996" type="CombineSegments" version="extension">
 			<input>
 				<port id="0" precision="I32">
 					<dim>-1</dim>
@@ -538,7 +538,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="39" name="Subtract_133713" type="Subtract" version="opset1">
+		<layer id="39" name="Subtract_134997" type="Subtract" version="opset1">
 			<data auto_broadcast="numpy" />
 			<input>
 				<port id="0" precision="I32">
@@ -554,13 +554,13 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="40" name="Constant_133714" type="Const" version="opset1">
+		<layer id="40" name="Constant_134998" type="Const" version="opset1">
 			<data element_type="i32" shape="" offset="1829035" size="4" />
 			<output>
 				<port id="0" precision="I32" />
 			</output>
 		</layer>
-		<layer id="41" name="ReduceMax_133715" type="ReduceMax" version="opset1">
+		<layer id="41" name="ReduceMax_134999" type="ReduceMax" version="opset1">
 			<data keep_dims="false" />
 			<input>
 				<port id="0" precision="I32">
@@ -572,13 +572,13 @@
 				<port id="2" precision="I32" />
 			</output>
 		</layer>
-		<layer id="42" name="Constant_133716" type="Const" version="opset1">
+		<layer id="42" name="Constant_135000" type="Const" version="opset1">
 			<data element_type="i32" shape="" offset="1829035" size="4" />
 			<output>
 				<port id="0" precision="I32" />
 			</output>
 		</layer>
-		<layer id="43" name="RaggedToDense_133717" type="RaggedToDense" version="extension">
+		<layer id="43" name="RaggedToDense_135001" type="RaggedToDense" version="extension">
 			<data pad_right="false" m_pad_max_length="false" />
 			<input>
 				<port id="0" precision="I32">
@@ -604,7 +604,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="44" name="Convert_133718" type="Convert" version="opset1">
+		<layer id="44" name="Convert_135002" type="Convert" version="opset1">
 			<data destination_type="i32" />
 			<input>
 				<port id="0" precision="BOOL">
@@ -619,7 +619,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="45" name="Convert_133718.0" type="Convert" version="opset1">
+		<layer id="45" name="Convert_135002.0" type="Convert" version="opset1">
 			<data destination_type="i64" />
 			<input>
 				<port id="0" precision="I32">
@@ -634,7 +634,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="47" name="RaggedToDense_133717.0" type="Convert" version="opset1">
+		<layer id="47" name="RaggedToDense_135001.0" type="Convert" version="opset1">
 			<data destination_type="i64" />
 			<input>
 				<port id="0" precision="I32">
@@ -649,7 +649,7 @@
 				</port>
 			</output>
 		</layer>
-		<layer id="48" name="Result_133721" type="Result" version="opset1" output_names="input_ids">
+		<layer id="48" name="Result_135005" type="Result" version="opset1" output_names="input_ids">
 			<input>
 				<port id="0" precision="I64">
 					<dim>-1</dim>
@@ -657,7 +657,7 @@
 				</port>
 			</input>
 		</layer>
-		<layer id="46" name="Result_133723" type="Result" version="opset1" output_names="attention_mask">
+		<layer id="46" name="Result_135007" type="Result" version="opset1" output_names="attention_mask">
 			<input>
 				<port id="0" precision="I64">
 					<dim>-1</dim>

tokenizer_config.json

@@ -207,6 +207,7 @@
   "bos_token": "<|endoftext|>",
   "clean_up_tokenization_spaces": true,
   "eos_token": "<|endoftext|>",
+  "extra_special_tokens": {},
   "model_max_length": 2048,
   "pad_token": null,
   "tokenizer_class": "GPTNeoXTokenizer",