Qalam_onnx / quant_config.json
gagan3012's picture
First model version
3db9897
{
"per_channel": true,
"reduce_range": true,
"per_model_config": {
"decoder_model": {
"op_types": [
"Constant",
"Concat",
"Pow",
"Cast",
"Slice",
"Sqrt",
"Equal",
"Transpose",
"Unsqueeze",
"MatMul",
"Div",
"Where",
"ConstantOfShape",
"Reshape",
"Shape",
"Mul",
"Add",
"CumSum",
"Softmax",
"Erf",
"Expand",
"Tile",
"Sub",
"Range",
"Not",
"Gather",
"Greater",
"ReduceMean"
],
"weight_type": "QInt8"
},
"encoder_model": {
"op_types": [
"Constant",
"Concat",
"Pow",
"Cast",
"Slice",
"Pad",
"ReduceSum",
"Sqrt",
"Equal",
"Transpose",
"Mod",
"Unsqueeze",
"MatMul",
"Div",
"Where",
"ConstantOfShape",
"Reshape",
"Abs",
"Shape",
"Mul",
"Exp",
"Add",
"Relu",
"Sigmoid",
"ScatterND",
"Softmax",
"Erf",
"Expand",
"Sub",
"Range",
"Not",
"Conv",
"Gather",
"Clip",
"ReduceMean"
],
"weight_type": "QUInt8"
}
}
}