quantize_config.json
749 B · 33 lines · json Raw
1 {
2 "per_channel": true,
3 "reduce_range": true,
4 "per_model_config": {
5 "model": {
6 "op_types": [
7 "Reshape",
8 "Sqrt",
9 "Unsqueeze",
10 "Pow",
11 "Mul",
12 "Slice",
13 "Gather",
14 "Sub",
15 "Relu",
16 "Resize",
17 "Softmax",
18 "ReduceMean",
19 "Add",
20 "MatMul",
21 "Erf",
22 "Div",
23 "Constant",
24 "Transpose",
25 "Conv",
26 "Shape",
27 "Cast",
28 "Concat"
29 ],
30 "weight_type": "QUInt8"
31 }
32 }
33 }