Vadim Borisov commited on
Upload quantize_config.json with huggingface_hub
Browse files- quantize_config.json +32 -0
quantize_config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"per_channel": true,
|
| 3 |
+
"reduce_range": true,
|
| 4 |
+
"per_model_config": {
|
| 5 |
+
"model": {
|
| 6 |
+
"op_types": [
|
| 7 |
+
"Reshape",
|
| 8 |
+
"Sqrt",
|
| 9 |
+
"Tanh",
|
| 10 |
+
"Mul",
|
| 11 |
+
"Concat",
|
| 12 |
+
"Add",
|
| 13 |
+
"ReduceMean",
|
| 14 |
+
"Cast",
|
| 15 |
+
"Sub",
|
| 16 |
+
"Erf",
|
| 17 |
+
"MatMul",
|
| 18 |
+
"Slice",
|
| 19 |
+
"Unsqueeze",
|
| 20 |
+
"Transpose",
|
| 21 |
+
"Constant",
|
| 22 |
+
"Gemm",
|
| 23 |
+
"Shape",
|
| 24 |
+
"Softmax",
|
| 25 |
+
"Gather",
|
| 26 |
+
"Div",
|
| 27 |
+
"Pow"
|
| 28 |
+
],
|
| 29 |
+
"weight_type": "QInt8"
|
| 30 |
+
}
|
| 31 |
+
}
|
| 32 |
+
}
|