initial commit

config.json

@@ -0,0 +1,68 @@
+{
+  "_name_or_path": "/tmp/tmp164lo86z",
+  "architectures": [
+    "BertForTokenClassification"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "classifier_dropout": null,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "id2label": {
+    "0": "LABEL_0",
+    "1": "LABEL_1",
+    "2": "LABEL_2",
+    "3": "LABEL_3",
+    "4": "LABEL_4",
+    "5": "LABEL_5",
+    "6": "LABEL_6",
+    "7": "LABEL_7",
+    "8": "LABEL_8",
+    "9": "LABEL_9",
+    "10": "LABEL_10",
+    "11": "LABEL_11",
+    "12": "LABEL_12",
+    "13": "LABEL_13",
+    "14": "LABEL_14",
+    "15": "LABEL_15",
+    "16": "LABEL_16",
+    "17": "LABEL_17",
+    "18": "LABEL_18"
+  },
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "label2id": {
+    "LABEL_0": 0,
+    "LABEL_1": 1,
+    "LABEL_10": 10,
+    "LABEL_11": 11,
+    "LABEL_12": 12,
+    "LABEL_13": 13,
+    "LABEL_14": 14,
+    "LABEL_15": 15,
+    "LABEL_16": 16,
+    "LABEL_17": 17,
+    "LABEL_18": 18,
+    "LABEL_2": 2,
+    "LABEL_3": 3,
+    "LABEL_4": 4,
+    "LABEL_5": 5,
+    "LABEL_6": 6,
+    "LABEL_7": 7,
+    "LABEL_8": 8,
+    "LABEL_9": 9
+  },
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 6,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "torch_dtype": "float32",
+  "transformers_version": "4.38.2",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "vocab_size": 29794
+}

model_quantized.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8850eceffe2666f6924be58d8138ae8c3f78ea9b1fdb96fe2f3f839529354301
+size 87519705

ort_config.json

@@ -0,0 +1,596 @@
+{
+  "one_external_file": true,
+  "opset": null,
+  "optimization": {},
+  "optimum_version": "1.18.0",
+  "quantization": {
+    "activations_dtype": "QUInt8",
+    "activations_symmetric": false,
+    "format": "QOperator",
+    "is_static": false,
+    "mode": "IntegerOps",
+    "nodes_to_exclude": [],
+    "nodes_to_quantize": [
+      "/bert/Constant",
+      "/bert/Constant_1",
+      "/bert/embeddings/Constant",
+      "Constant_157",
+      "/bert/embeddings/Constant_1",
+      "/bert/embeddings/Constant_2",
+      "/bert/embeddings/Constant_3",
+      "/bert/embeddings/LayerNorm/Constant",
+      "/bert/embeddings/LayerNorm/Constant_1",
+      "/bert/encoder/layer.0/attention/self/Constant",
+      "/bert/encoder/layer.0/attention/self/Constant_1",
+      "/bert/encoder/layer.0/attention/self/Constant_2",
+      "/bert/encoder/layer.0/attention/self/Constant_3",
+      "/bert/encoder/layer.0/attention/self/Constant_4",
+      "/bert/encoder/layer.0/attention/self/Constant_5",
+      "/bert/encoder/layer.0/attention/self/Constant_6",
+      "/bert/encoder/layer.0/attention/self/Constant_7",
+      "/bert/encoder/layer.0/attention/self/Constant_8",
+      "/bert/encoder/layer.0/attention/self/Constant_9",
+      "/bert/encoder/layer.0/attention/self/Constant_10",
+      "/bert/encoder/layer.0/attention/self/Constant_11",
+      "/bert/encoder/layer.0/attention/self/Constant_12",
+      "/bert/encoder/layer.0/attention/self/Constant_13",
+      "/bert/encoder/layer.0/attention/self/Constant_14",
+      "/bert/encoder/layer.0/attention/self/Constant_15",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Constant",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Constant",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Constant_1",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Constant_2",
+      "/bert/encoder/layer.0/output/LayerNorm/Constant",
+      "/bert/encoder/layer.0/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.1/attention/self/Constant",
+      "/bert/encoder/layer.1/attention/self/Constant_1",
+      "/bert/encoder/layer.1/attention/self/Constant_2",
+      "/bert/encoder/layer.1/attention/self/Constant_3",
+      "/bert/encoder/layer.1/attention/self/Constant_4",
+      "/bert/encoder/layer.1/attention/self/Constant_5",
+      "/bert/encoder/layer.1/attention/self/Constant_6",
+      "/bert/encoder/layer.1/attention/self/Constant_7",
+      "/bert/encoder/layer.1/attention/self/Constant_8",
+      "/bert/encoder/layer.1/attention/self/Constant_9",
+      "/bert/encoder/layer.1/attention/self/Constant_10",
+      "/bert/encoder/layer.1/attention/self/Constant_11",
+      "/bert/encoder/layer.1/attention/self/Constant_12",
+      "/bert/encoder/layer.1/attention/self/Constant_13",
+      "/bert/encoder/layer.1/attention/self/Constant_14",
+      "/bert/encoder/layer.1/attention/self/Constant_15",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Constant",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Constant",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Constant_1",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Constant_2",
+      "/bert/encoder/layer.1/output/LayerNorm/Constant",
+      "/bert/encoder/layer.1/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.2/attention/self/Constant",
+      "/bert/encoder/layer.2/attention/self/Constant_1",
+      "/bert/encoder/layer.2/attention/self/Constant_2",
+      "/bert/encoder/layer.2/attention/self/Constant_3",
+      "/bert/encoder/layer.2/attention/self/Constant_4",
+      "/bert/encoder/layer.2/attention/self/Constant_5",
+      "/bert/encoder/layer.2/attention/self/Constant_6",
+      "/bert/encoder/layer.2/attention/self/Constant_7",
+      "/bert/encoder/layer.2/attention/self/Constant_8",
+      "/bert/encoder/layer.2/attention/self/Constant_9",
+      "/bert/encoder/layer.2/attention/self/Constant_10",
+      "/bert/encoder/layer.2/attention/self/Constant_11",
+      "/bert/encoder/layer.2/attention/self/Constant_12",
+      "/bert/encoder/layer.2/attention/self/Constant_13",
+      "/bert/encoder/layer.2/attention/self/Constant_14",
+      "/bert/encoder/layer.2/attention/self/Constant_15",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Constant",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Constant",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Constant_1",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Constant_2",
+      "/bert/encoder/layer.2/output/LayerNorm/Constant",
+      "/bert/encoder/layer.2/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.3/attention/self/Constant",
+      "/bert/encoder/layer.3/attention/self/Constant_1",
+      "/bert/encoder/layer.3/attention/self/Constant_2",
+      "/bert/encoder/layer.3/attention/self/Constant_3",
+      "/bert/encoder/layer.3/attention/self/Constant_4",
+      "/bert/encoder/layer.3/attention/self/Constant_5",
+      "/bert/encoder/layer.3/attention/self/Constant_6",
+      "/bert/encoder/layer.3/attention/self/Constant_7",
+      "/bert/encoder/layer.3/attention/self/Constant_8",
+      "/bert/encoder/layer.3/attention/self/Constant_9",
+      "/bert/encoder/layer.3/attention/self/Constant_10",
+      "/bert/encoder/layer.3/attention/self/Constant_11",
+      "/bert/encoder/layer.3/attention/self/Constant_12",
+      "/bert/encoder/layer.3/attention/self/Constant_13",
+      "/bert/encoder/layer.3/attention/self/Constant_14",
+      "/bert/encoder/layer.3/attention/self/Constant_15",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Constant",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Constant",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Constant_1",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Constant_2",
+      "/bert/encoder/layer.3/output/LayerNorm/Constant",
+      "/bert/encoder/layer.3/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.4/attention/self/Constant",
+      "/bert/encoder/layer.4/attention/self/Constant_1",
+      "/bert/encoder/layer.4/attention/self/Constant_2",
+      "/bert/encoder/layer.4/attention/self/Constant_3",
+      "/bert/encoder/layer.4/attention/self/Constant_4",
+      "/bert/encoder/layer.4/attention/self/Constant_5",
+      "/bert/encoder/layer.4/attention/self/Constant_6",
+      "/bert/encoder/layer.4/attention/self/Constant_7",
+      "/bert/encoder/layer.4/attention/self/Constant_8",
+      "/bert/encoder/layer.4/attention/self/Constant_9",
+      "/bert/encoder/layer.4/attention/self/Constant_10",
+      "/bert/encoder/layer.4/attention/self/Constant_11",
+      "/bert/encoder/layer.4/attention/self/Constant_12",
+      "/bert/encoder/layer.4/attention/self/Constant_13",
+      "/bert/encoder/layer.4/attention/self/Constant_14",
+      "/bert/encoder/layer.4/attention/self/Constant_15",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Constant",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Constant",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Constant_1",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Constant_2",
+      "/bert/encoder/layer.4/output/LayerNorm/Constant",
+      "/bert/encoder/layer.4/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.5/attention/self/Constant",
+      "/bert/encoder/layer.5/attention/self/Constant_1",
+      "/bert/encoder/layer.5/attention/self/Constant_2",
+      "/bert/encoder/layer.5/attention/self/Constant_3",
+      "/bert/encoder/layer.5/attention/self/Constant_4",
+      "/bert/encoder/layer.5/attention/self/Constant_5",
+      "/bert/encoder/layer.5/attention/self/Constant_6",
+      "/bert/encoder/layer.5/attention/self/Constant_7",
+      "/bert/encoder/layer.5/attention/self/Constant_8",
+      "/bert/encoder/layer.5/attention/self/Constant_9",
+      "/bert/encoder/layer.5/attention/self/Constant_10",
+      "/bert/encoder/layer.5/attention/self/Constant_11",
+      "/bert/encoder/layer.5/attention/self/Constant_12",
+      "/bert/encoder/layer.5/attention/self/Constant_13",
+      "/bert/encoder/layer.5/attention/self/Constant_14",
+      "/bert/encoder/layer.5/attention/self/Constant_15",
+      "/bert/encoder/layer.5/attention/output/LayerNorm/Constant",
+      "/bert/encoder/layer.5/attention/output/LayerNorm/Constant_1",
+      "/bert/encoder/layer.5/intermediate/intermediate_act_fn/Constant",
+      "/bert/encoder/layer.5/intermediate/intermediate_act_fn/Constant_1",
+      "/bert/encoder/layer.5/intermediate/intermediate_act_fn/Constant_2",
+      "/bert/encoder/layer.5/output/LayerNorm/Constant",
+      "/bert/encoder/layer.5/output/LayerNorm/Constant_1",
+      "/bert/Unsqueeze",
+      "/bert/embeddings/Shape",
+      "/bert/embeddings/word_embeddings/Gather",
+      "/bert/embeddings/token_type_embeddings/Gather",
+      "/bert/Unsqueeze_1",
+      "/bert/embeddings/Gather",
+      "/bert/embeddings/Add",
+      "/bert/Cast",
+      "/bert/embeddings/Unsqueeze",
+      "/bert/Sub",
+      "/bert/embeddings/Slice",
+      "/bert/Mul",
+      "/bert/embeddings/position_embeddings/Gather",
+      "/bert/embeddings/Add_1",
+      "/bert/embeddings/LayerNorm/ReduceMean",
+      "/bert/embeddings/LayerNorm/Sub",
+      "/bert/embeddings/LayerNorm/Pow",
+      "/bert/embeddings/LayerNorm/ReduceMean_1",
+      "/bert/embeddings/LayerNorm/Add",
+      "/bert/embeddings/LayerNorm/Sqrt",
+      "/bert/embeddings/LayerNorm/Div",
+      "/bert/embeddings/LayerNorm/Mul",
+      "/bert/embeddings/LayerNorm/Add_1",
+      "/bert/encoder/layer.0/attention/self/query/MatMul",
+      "/bert/encoder/layer.0/attention/self/key/MatMul",
+      "/bert/encoder/layer.0/attention/self/value/MatMul",
+      "/bert/encoder/layer.0/attention/self/query/Add",
+      "/bert/encoder/layer.0/attention/self/key/Add",
+      "/bert/encoder/layer.0/attention/self/value/Add",
+      "/bert/encoder/layer.0/attention/self/Shape_4",
+      "/bert/encoder/layer.0/attention/self/Shape_5",
+      "/bert/encoder/layer.0/attention/self/Shape",
+      "/bert/encoder/layer.0/attention/self/Shape_1",
+      "/bert/encoder/layer.0/attention/self/Shape_2",
+      "/bert/encoder/layer.0/attention/self/Shape_3",
+      "/bert/encoder/layer.0/attention/self/Gather_4",
+      "/bert/encoder/layer.0/attention/self/Gather_5",
+      "/bert/encoder/layer.0/attention/self/Gather",
+      "/bert/encoder/layer.0/attention/self/Gather_1",
+      "/bert/encoder/layer.0/attention/self/Gather_2",
+      "/bert/encoder/layer.0/attention/self/Gather_3",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze_4",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze_5",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze_1",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze_2",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze_3",
+      "/bert/encoder/layer.0/attention/self/Concat_2",
+      "/bert/encoder/layer.0/attention/self/Concat",
+      "/bert/encoder/layer.0/attention/self/Concat_1",
+      "/bert/encoder/layer.0/attention/self/Reshape_2",
+      "/bert/encoder/layer.0/attention/self/Reshape",
+      "/bert/encoder/layer.0/attention/self/Reshape_1",
+      "/bert/encoder/layer.0/attention/self/Transpose_1",
+      "/bert/encoder/layer.0/attention/self/Transpose_2",
+      "/bert/encoder/layer.0/attention/self/Transpose",
+      "/bert/encoder/layer.0/attention/self/MatMul",
+      "/bert/encoder/layer.0/attention/self/Div",
+      "/bert/encoder/layer.0/attention/self/Add",
+      "/bert/encoder/layer.0/attention/self/Softmax",
+      "/bert/encoder/layer.0/attention/self/MatMul_1",
+      "/bert/encoder/layer.0/attention/self/Transpose_3",
+      "/bert/encoder/layer.0/attention/self/Shape_6",
+      "/bert/encoder/layer.0/attention/self/Shape_7",
+      "/bert/encoder/layer.0/attention/self/Gather_6",
+      "/bert/encoder/layer.0/attention/self/Gather_7",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze_6",
+      "/bert/encoder/layer.0/attention/self/Unsqueeze_7",
+      "/bert/encoder/layer.0/attention/self/Concat_3",
+      "/bert/encoder/layer.0/attention/self/Reshape_3",
+      "/bert/encoder/layer.0/attention/output/dense/MatMul",
+      "/bert/encoder/layer.0/attention/output/dense/Add",
+      "/bert/encoder/layer.0/attention/output/Add",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Sub",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Pow",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Add",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Div",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Mul",
+      "/bert/encoder/layer.0/attention/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.0/intermediate/dense/MatMul",
+      "/bert/encoder/layer.0/intermediate/dense/Add",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Div",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Erf",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Add",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Mul",
+      "/bert/encoder/layer.0/intermediate/intermediate_act_fn/Mul_1",
+      "/bert/encoder/layer.0/output/dense/MatMul",
+      "/bert/encoder/layer.0/output/dense/Add",
+      "/bert/encoder/layer.0/output/Add",
+      "/bert/encoder/layer.0/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.0/output/LayerNorm/Sub",
+      "/bert/encoder/layer.0/output/LayerNorm/Pow",
+      "/bert/encoder/layer.0/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.0/output/LayerNorm/Add",
+      "/bert/encoder/layer.0/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.0/output/LayerNorm/Div",
+      "/bert/encoder/layer.0/output/LayerNorm/Mul",
+      "/bert/encoder/layer.0/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.1/attention/self/query/MatMul",
+      "/bert/encoder/layer.1/attention/self/key/MatMul",
+      "/bert/encoder/layer.1/attention/self/value/MatMul",
+      "/bert/encoder/layer.1/attention/self/query/Add",
+      "/bert/encoder/layer.1/attention/self/key/Add",
+      "/bert/encoder/layer.1/attention/self/value/Add",
+      "/bert/encoder/layer.1/attention/self/Shape_4",
+      "/bert/encoder/layer.1/attention/self/Shape_5",
+      "/bert/encoder/layer.1/attention/self/Shape",
+      "/bert/encoder/layer.1/attention/self/Shape_1",
+      "/bert/encoder/layer.1/attention/self/Shape_2",
+      "/bert/encoder/layer.1/attention/self/Shape_3",
+      "/bert/encoder/layer.1/attention/self/Gather_4",
+      "/bert/encoder/layer.1/attention/self/Gather_5",
+      "/bert/encoder/layer.1/attention/self/Gather",
+      "/bert/encoder/layer.1/attention/self/Gather_1",
+      "/bert/encoder/layer.1/attention/self/Gather_2",
+      "/bert/encoder/layer.1/attention/self/Gather_3",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze_4",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze_5",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze_1",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze_2",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze_3",
+      "/bert/encoder/layer.1/attention/self/Concat_2",
+      "/bert/encoder/layer.1/attention/self/Concat",
+      "/bert/encoder/layer.1/attention/self/Concat_1",
+      "/bert/encoder/layer.1/attention/self/Reshape_2",
+      "/bert/encoder/layer.1/attention/self/Reshape",
+      "/bert/encoder/layer.1/attention/self/Reshape_1",
+      "/bert/encoder/layer.1/attention/self/Transpose_1",
+      "/bert/encoder/layer.1/attention/self/Transpose_2",
+      "/bert/encoder/layer.1/attention/self/Transpose",
+      "/bert/encoder/layer.1/attention/self/MatMul",
+      "/bert/encoder/layer.1/attention/self/Div",
+      "/bert/encoder/layer.1/attention/self/Add",
+      "/bert/encoder/layer.1/attention/self/Softmax",
+      "/bert/encoder/layer.1/attention/self/MatMul_1",
+      "/bert/encoder/layer.1/attention/self/Transpose_3",
+      "/bert/encoder/layer.1/attention/self/Shape_6",
+      "/bert/encoder/layer.1/attention/self/Shape_7",
+      "/bert/encoder/layer.1/attention/self/Gather_6",
+      "/bert/encoder/layer.1/attention/self/Gather_7",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze_6",
+      "/bert/encoder/layer.1/attention/self/Unsqueeze_7",
+      "/bert/encoder/layer.1/attention/self/Concat_3",
+      "/bert/encoder/layer.1/attention/self/Reshape_3",
+      "/bert/encoder/layer.1/attention/output/dense/MatMul",
+      "/bert/encoder/layer.1/attention/output/dense/Add",
+      "/bert/encoder/layer.1/attention/output/Add",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Sub",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Pow",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Add",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Div",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Mul",
+      "/bert/encoder/layer.1/attention/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.1/intermediate/dense/MatMul",
+      "/bert/encoder/layer.1/intermediate/dense/Add",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Div",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Erf",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Add",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Mul",
+      "/bert/encoder/layer.1/intermediate/intermediate_act_fn/Mul_1",
+      "/bert/encoder/layer.1/output/dense/MatMul",
+      "/bert/encoder/layer.1/output/dense/Add",
+      "/bert/encoder/layer.1/output/Add",
+      "/bert/encoder/layer.1/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.1/output/LayerNorm/Sub",
+      "/bert/encoder/layer.1/output/LayerNorm/Pow",
+      "/bert/encoder/layer.1/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.1/output/LayerNorm/Add",
+      "/bert/encoder/layer.1/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.1/output/LayerNorm/Div",
+      "/bert/encoder/layer.1/output/LayerNorm/Mul",
+      "/bert/encoder/layer.1/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.2/attention/self/query/MatMul",
+      "/bert/encoder/layer.2/attention/self/key/MatMul",
+      "/bert/encoder/layer.2/attention/self/value/MatMul",
+      "/bert/encoder/layer.2/attention/self/query/Add",
+      "/bert/encoder/layer.2/attention/self/key/Add",
+      "/bert/encoder/layer.2/attention/self/value/Add",
+      "/bert/encoder/layer.2/attention/self/Shape_4",
+      "/bert/encoder/layer.2/attention/self/Shape_5",
+      "/bert/encoder/layer.2/attention/self/Shape",
+      "/bert/encoder/layer.2/attention/self/Shape_1",
+      "/bert/encoder/layer.2/attention/self/Shape_2",
+      "/bert/encoder/layer.2/attention/self/Shape_3",
+      "/bert/encoder/layer.2/attention/self/Gather_4",
+      "/bert/encoder/layer.2/attention/self/Gather_5",
+      "/bert/encoder/layer.2/attention/self/Gather",
+      "/bert/encoder/layer.2/attention/self/Gather_1",
+      "/bert/encoder/layer.2/attention/self/Gather_2",
+      "/bert/encoder/layer.2/attention/self/Gather_3",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze_4",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze_5",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze_1",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze_2",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze_3",
+      "/bert/encoder/layer.2/attention/self/Concat_2",
+      "/bert/encoder/layer.2/attention/self/Concat",
+      "/bert/encoder/layer.2/attention/self/Concat_1",
+      "/bert/encoder/layer.2/attention/self/Reshape_2",
+      "/bert/encoder/layer.2/attention/self/Reshape",
+      "/bert/encoder/layer.2/attention/self/Reshape_1",
+      "/bert/encoder/layer.2/attention/self/Transpose_1",
+      "/bert/encoder/layer.2/attention/self/Transpose_2",
+      "/bert/encoder/layer.2/attention/self/Transpose",
+      "/bert/encoder/layer.2/attention/self/MatMul",
+      "/bert/encoder/layer.2/attention/self/Div",
+      "/bert/encoder/layer.2/attention/self/Add",
+      "/bert/encoder/layer.2/attention/self/Softmax",
+      "/bert/encoder/layer.2/attention/self/MatMul_1",
+      "/bert/encoder/layer.2/attention/self/Transpose_3",
+      "/bert/encoder/layer.2/attention/self/Shape_6",
+      "/bert/encoder/layer.2/attention/self/Shape_7",
+      "/bert/encoder/layer.2/attention/self/Gather_6",
+      "/bert/encoder/layer.2/attention/self/Gather_7",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze_6",
+      "/bert/encoder/layer.2/attention/self/Unsqueeze_7",
+      "/bert/encoder/layer.2/attention/self/Concat_3",
+      "/bert/encoder/layer.2/attention/self/Reshape_3",
+      "/bert/encoder/layer.2/attention/output/dense/MatMul",
+      "/bert/encoder/layer.2/attention/output/dense/Add",
+      "/bert/encoder/layer.2/attention/output/Add",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Sub",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Pow",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Add",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Div",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Mul",
+      "/bert/encoder/layer.2/attention/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.2/intermediate/dense/MatMul",
+      "/bert/encoder/layer.2/intermediate/dense/Add",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Div",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Erf",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Add",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Mul",
+      "/bert/encoder/layer.2/intermediate/intermediate_act_fn/Mul_1",
+      "/bert/encoder/layer.2/output/dense/MatMul",
+      "/bert/encoder/layer.2/output/dense/Add",
+      "/bert/encoder/layer.2/output/Add",
+      "/bert/encoder/layer.2/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.2/output/LayerNorm/Sub",
+      "/bert/encoder/layer.2/output/LayerNorm/Pow",
+      "/bert/encoder/layer.2/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.2/output/LayerNorm/Add",
+      "/bert/encoder/layer.2/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.2/output/LayerNorm/Div",
+      "/bert/encoder/layer.2/output/LayerNorm/Mul",
+      "/bert/encoder/layer.2/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.3/attention/self/query/MatMul",
+      "/bert/encoder/layer.3/attention/self/key/MatMul",
+      "/bert/encoder/layer.3/attention/self/value/MatMul",
+      "/bert/encoder/layer.3/attention/self/query/Add",
+      "/bert/encoder/layer.3/attention/self/key/Add",
+      "/bert/encoder/layer.3/attention/self/value/Add",
+      "/bert/encoder/layer.3/attention/self/Shape_4",
+      "/bert/encoder/layer.3/attention/self/Shape_5",
+      "/bert/encoder/layer.3/attention/self/Shape",
+      "/bert/encoder/layer.3/attention/self/Shape_1",
+      "/bert/encoder/layer.3/attention/self/Shape_2",
+      "/bert/encoder/layer.3/attention/self/Shape_3",
+      "/bert/encoder/layer.3/attention/self/Gather_4",
+      "/bert/encoder/layer.3/attention/self/Gather_5",
+      "/bert/encoder/layer.3/attention/self/Gather",
+      "/bert/encoder/layer.3/attention/self/Gather_1",
+      "/bert/encoder/layer.3/attention/self/Gather_2",
+      "/bert/encoder/layer.3/attention/self/Gather_3",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze_4",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze_5",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze_1",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze_2",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze_3",
+      "/bert/encoder/layer.3/attention/self/Concat_2",
+      "/bert/encoder/layer.3/attention/self/Concat",
+      "/bert/encoder/layer.3/attention/self/Concat_1",
+      "/bert/encoder/layer.3/attention/self/Reshape_2",
+      "/bert/encoder/layer.3/attention/self/Reshape",
+      "/bert/encoder/layer.3/attention/self/Reshape_1",
+      "/bert/encoder/layer.3/attention/self/Transpose_1",
+      "/bert/encoder/layer.3/attention/self/Transpose_2",
+      "/bert/encoder/layer.3/attention/self/Transpose",
+      "/bert/encoder/layer.3/attention/self/MatMul",
+      "/bert/encoder/layer.3/attention/self/Div",
+      "/bert/encoder/layer.3/attention/self/Add",
+      "/bert/encoder/layer.3/attention/self/Softmax",
+      "/bert/encoder/layer.3/attention/self/MatMul_1",
+      "/bert/encoder/layer.3/attention/self/Transpose_3",
+      "/bert/encoder/layer.3/attention/self/Shape_6",
+      "/bert/encoder/layer.3/attention/self/Shape_7",
+      "/bert/encoder/layer.3/attention/self/Gather_6",
+      "/bert/encoder/layer.3/attention/self/Gather_7",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze_6",
+      "/bert/encoder/layer.3/attention/self/Unsqueeze_7",
+      "/bert/encoder/layer.3/attention/self/Concat_3",
+      "/bert/encoder/layer.3/attention/self/Reshape_3",
+      "/bert/encoder/layer.3/attention/output/dense/MatMul",
+      "/bert/encoder/layer.3/attention/output/dense/Add",
+      "/bert/encoder/layer.3/attention/output/Add",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Sub",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Pow",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Add",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Div",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Mul",
+      "/bert/encoder/layer.3/attention/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.3/intermediate/dense/MatMul",
+      "/bert/encoder/layer.3/intermediate/dense/Add",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Div",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Erf",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Add",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Mul",
+      "/bert/encoder/layer.3/intermediate/intermediate_act_fn/Mul_1",
+      "/bert/encoder/layer.3/output/dense/MatMul",
+      "/bert/encoder/layer.3/output/dense/Add",
+      "/bert/encoder/layer.3/output/Add",
+      "/bert/encoder/layer.3/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.3/output/LayerNorm/Sub",
+      "/bert/encoder/layer.3/output/LayerNorm/Pow",
+      "/bert/encoder/layer.3/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.3/output/LayerNorm/Add",
+      "/bert/encoder/layer.3/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.3/output/LayerNorm/Div",
+      "/bert/encoder/layer.3/output/LayerNorm/Mul",
+      "/bert/encoder/layer.3/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.4/attention/self/query/MatMul",
+      "/bert/encoder/layer.4/attention/self/key/MatMul",
+      "/bert/encoder/layer.4/attention/self/value/MatMul",
+      "/bert/encoder/layer.4/attention/self/query/Add",
+      "/bert/encoder/layer.4/attention/self/key/Add",
+      "/bert/encoder/layer.4/attention/self/value/Add",
+      "/bert/encoder/layer.4/attention/self/Shape_4",
+      "/bert/encoder/layer.4/attention/self/Shape_5",
+      "/bert/encoder/layer.4/attention/self/Shape",
+      "/bert/encoder/layer.4/attention/self/Shape_1",
+      "/bert/encoder/layer.4/attention/self/Shape_2",
+      "/bert/encoder/layer.4/attention/self/Shape_3",
+      "/bert/encoder/layer.4/attention/self/Gather_4",
+      "/bert/encoder/layer.4/attention/self/Gather_5",
+      "/bert/encoder/layer.4/attention/self/Gather",
+      "/bert/encoder/layer.4/attention/self/Gather_1",
+      "/bert/encoder/layer.4/attention/self/Gather_2",
+      "/bert/encoder/layer.4/attention/self/Gather_3",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze_4",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze_5",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze_1",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze_2",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze_3",
+      "/bert/encoder/layer.4/attention/self/Concat_2",
+      "/bert/encoder/layer.4/attention/self/Concat",
+      "/bert/encoder/layer.4/attention/self/Concat_1",
+      "/bert/encoder/layer.4/attention/self/Reshape_2",
+      "/bert/encoder/layer.4/attention/self/Reshape",
+      "/bert/encoder/layer.4/attention/self/Reshape_1",
+      "/bert/encoder/layer.4/attention/self/Transpose_1",
+      "/bert/encoder/layer.4/attention/self/Transpose_2",
+      "/bert/encoder/layer.4/attention/self/Transpose",
+      "/bert/encoder/layer.4/attention/self/MatMul",
+      "/bert/encoder/layer.4/attention/self/Div",
+      "/bert/encoder/layer.4/attention/self/Add",
+      "/bert/encoder/layer.4/attention/self/Softmax",
+      "/bert/encoder/layer.4/attention/self/MatMul_1",
+      "/bert/encoder/layer.4/attention/self/Transpose_3",
+      "/bert/encoder/layer.4/attention/self/Shape_6",
+      "/bert/encoder/layer.4/attention/self/Shape_7",
+      "/bert/encoder/layer.4/attention/self/Gather_6",
+      "/bert/encoder/layer.4/attention/self/Gather_7",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze_6",
+      "/bert/encoder/layer.4/attention/self/Unsqueeze_7",
+      "/bert/encoder/layer.4/attention/self/Concat_3",
+      "/bert/encoder/layer.4/attention/self/Reshape_3",
+      "/bert/encoder/layer.4/attention/output/dense/MatMul",
+      "/bert/encoder/layer.4/attention/output/dense/Add",
+      "/bert/encoder/layer.4/attention/output/Add",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Sub",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Pow",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Add",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Div",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Mul",
+      "/bert/encoder/layer.4/attention/output/LayerNorm/Add_1",
+      "/bert/encoder/layer.4/intermediate/dense/MatMul",
+      "/bert/encoder/layer.4/intermediate/dense/Add",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Div",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Erf",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Add",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Mul",
+      "/bert/encoder/layer.4/intermediate/intermediate_act_fn/Mul_1",
+      "/bert/encoder/layer.4/output/dense/MatMul",
+      "/bert/encoder/layer.4/output/dense/Add",
+      "/bert/encoder/layer.4/output/Add",
+      "/bert/encoder/layer.4/output/LayerNorm/ReduceMean",
+      "/bert/encoder/layer.4/output/LayerNorm/Sub",
+      "/bert/encoder/layer.4/output/LayerNorm/Pow",
+      "/bert/encoder/layer.4/output/LayerNorm/ReduceMean_1",
+      "/bert/encoder/layer.4/output/LayerNorm/Add",
+      "/bert/encoder/layer.4/output/LayerNorm/Sqrt",
+      "/bert/encoder/layer.4/output/LayerNorm/Div",
+      "/bert/encoder/layer.4/output/LayerNorm/Mul",
+      "/bert/encoder/layer.4/output/LayerNorm/Add_1"
+    ],
+    "operators_to_quantize": [
+      "Conv",
+      "MatMul",
+      "Attention",
+      "LSTM",
+      "Gather",
+      "Transpose",
+      "EmbedLayerNormalization"
+    ],
+    "per_channel": false,
+    "qdq_add_pair_to_weight": false,
+    "qdq_dedicated_pair": false,
+    "qdq_op_type_per_channel_support_to_axis": {
+      "MatMul": 1
+    },
+    "reduce_range": false,
+    "weights_dtype": "QInt8",
+    "weights_symmetric": true
+  },
+  "transformers_version": "4.38.2",
+  "use_external_data_format": false
+}

special_tokens_map.json

@@ -0,0 +1,37 @@
+{
+  "cls_token": {
+    "content": "[CLS]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "mask_token": {
+    "content": "[MASK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "[PAD]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "sep_token": {
+    "content": "[SEP]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "unk_token": {
+    "content": "[UNK]",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

tokenizer_config.json

@@ -0,0 +1,61 @@
+{
+  "added_tokens_decoder": {
+    "0": {
+      "content": "[PAD]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "100": {
+      "content": "[UNK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "101": {
+      "content": "[CLS]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "102": {
+      "content": "[SEP]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "103": {
+      "content": "[MASK]",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    }
+  },
+  "clean_up_tokenization_spaces": true,
+  "cls_token": "[CLS]",
+  "do_basic_tokenize": true,
+  "do_lower_case": false,
+  "mask_token": "[MASK]",
+  "max_length": 512,
+  "model_max_length": 1000000000000000019884624838656,
+  "never_split": null,
+  "pad_token": "[PAD]",
+  "sep_token": "[SEP]",
+  "stride": 0,
+  "strip_accents": null,
+  "tokenize_chinese_chars": true,
+  "tokenizer_class": "BertTokenizer",
+  "truncation_side": "right",
+  "truncation_strategy": "longest_first",
+  "unk_token": "[UNK]"
+}