Upload folder using huggingface_hub

config.json

@@ -44,7 +44,7 @@
   "output_past": false,
   "pad_token_id": 1,
   "scale_embedding": false,
-  "transformers_version": "4.29.2",
+  "transformers_version": "4.34.0.dev0",
   "use_cache": true,
   "vocab_size": 50265
 }

quantize_config.json

@@ -0,0 +1,42 @@
+{
+    "per_channel": true,
+    "reduce_range": true,
+    "per_model_config": {
+        "model": {
+            "op_types": [
+                "Erf",
+                "NonZero",
+                "Tanh",
+                "ReduceMean",
+                "Reshape",
+                "Transpose",
+                "Equal",
+                "Pow",
+                "Slice",
+                "Where",
+                "ScatterND",
+                "Expand",
+                "Shape",
+                "Concat",
+                "MatMul",
+                "Sqrt",
+                "Range",
+                "Less",
+                "Add",
+                "Gather",
+                "Unsqueeze",
+                "Div",
+                "Gemm",
+                "Constant",
+                "Mul",
+                "Softmax",
+                "Squeeze",
+                "ConstantOfShape",
+                "Cast",
+                "GatherND",
+                "Sub"
+            ],
+            "weight_type": "QInt8"
+        }
+    }
+}