quantize_config.json
674 B · 30 lines · json Raw
1 {
2 "per_channel": true,
3 "reduce_range": true,
4 "per_model_config": {
5 "model": {
6 "op_types": [
7 "Cast",
8 "Softmax",
9 "Sub",
10 "Shape",
11 "Constant",
12 "Transpose",
13 "Concat",
14 "ReduceMean",
15 "Gather",
16 "Pow",
17 "Mul",
18 "Sqrt",
19 "Reshape",
20 "Slice",
21 "Erf",
22 "MatMul",
23 "Div",
24 "Unsqueeze",
25 "Add"
26 ],
27 "weight_type": "QInt8"
28 }
29 }
30 }