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 "Unsqueeze",
8 "Shape",
9 "Transpose",
10 "Sqrt",
11 "Gather",
12 "Slice",
13 "Erf",
14 "Div",
15 "Reshape",
16 "Add",
17 "Cast",
18 "Sub",
19 "Concat",
20 "ReduceMean",
21 "Mul",
22 "Conv",
23 "Constant",
24 "Resize",
25 "Softmax",
26 "Pow",
27 "Relu",
28 "MatMul"
29 ],
30 "weight_type": "QUInt8"
31 }
32 }
33 }