james_says / quantize_config.json
cosmictoby's picture
onnx attmpt 1000
8662a91
raw
history blame
No virus
846 Bytes
{
"per_channel": false,
"reduce_range": false,
"per_model_config": {
"decoder_model": {
"op_types": [
"Range",
"Transpose",
"Softmax",
"MatMul",
"Tanh",
"Gather",
"ConstantOfShape",
"Add",
"Where",
"Shape",
"Cast",
"Squeeze",
"Unsqueeze",
"Div",
"Reshape",
"Slice",
"Sqrt",
"Constant",
"Pow",
"Gemm",
"Concat",
"Sub",
"ReduceMean",
"Mul",
"Split"
],
"weight_type": "QInt8"
}
}
}