Upload folder using huggingface_hub

config.json

@@ -20,7 +20,6 @@
   "output_past": true,
   "pad_token_id": 1,
   "position_embedding_type": "absolute",
-  "torch_dtype": "float32",
   "transformers_version": "4.37.2",
   "type_vocab_size": 1,
   "use_cache": true,

onnx/model.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5d89a0010dd39aa2cfa8b22bb49f06904c5bbf5877135f877da419480f40cde3
+size 607298

onnx/model.onnx_data

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1eebfb28493f67bba03ce0ef64bfdc7fc5a3bd9d7493f818bb1d78cd798416b4
+size 2266820608

onnx/model_quantized.onnx

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0826f8c1ab9edf1801db86c61919d4d108e8bfc0b809ec823ad366882ff0b77d
+size 569694530

quantize_config.json

@@ -4,34 +4,31 @@
     "per_model_config": {
         "model": {
             "op_types": [
-                "Concat",
+                "ReduceMean",
+                "Where",
+                "MatMul",
+                "Reshape",
+                "Gather",
+                "Cast",
                 "ConstantOfShape",
-                "Erf",
-                "Mul",
-                "Shape",
                 "Pow",
-                "Cast",
-                "Sqrt",
-                "Add",
-                "Gather",
-                "Not",
-                "CumSum",
-                "Abs",
-                "Slice",
-                "Where",
-                "Transpose",
-                "Div",
                 "Softmax",
-                "Reshape",
+                "Mul",
+                "Erf",
                 "Sub",
-                "MatMul",
-                "ReduceMean",
-                "Constant",
-                "Clip",
-                "Expand",
                 "Unsqueeze",
+                "Div",
+                "Sqrt",
+                "Expand",
+                "Add",
+                "Transpose",
+                "Shape",
                 "Equal",
-                "ReduceSum"
+                "Not",
+                "Constant",
+                "Slice",
+                "CumSum",
+                "Concat"
             ],
             "weight_type": "QInt8"
         }