whisper-small-dsntt1-tr-onnx / quantize_config.json
ysdede's picture
Upload 19 files
ad6703d verified
{
"per_channel": false,
"reduce_range": false,
"per_model_config": {
"decoder_model": {
"op_types": [
"Less",
"Expand",
"Where",
"ConstantOfShape",
"Constant",
"Softmax",
"Div",
"Erf",
"Pow",
"Unsqueeze",
"Equal",
"Transpose",
"ReduceMean",
"Cast",
"Squeeze",
"Range",
"Add",
"Sub",
"Concat",
"Gather",
"MatMul",
"Reshape",
"Mul",
"Slice",
"Shape",
"Sqrt"
],
"weight_type": "QInt8"
},
"decoder_model_merged": {
"op_types": [
"Less",
"Expand",
"Where",
"ConstantOfShape",
"Constant",
"Softmax",
"Div",
"Erf",
"If",
"Pow",
"Unsqueeze",
"Equal",
"Transpose",
"ReduceMean",
"Cast",
"Squeeze",
"Range",
"Add",
"Sub",
"Concat",
"Gather",
"MatMul",
"Reshape",
"Mul",
"Slice",
"Shape",
"Sqrt"
],
"weight_type": "QInt8"
},
"decoder_with_past_model": {
"op_types": [
"Div",
"Slice",
"Erf",
"Add",
"Shape",
"Sub",
"Concat",
"Gather",
"Pow",
"Sqrt",
"Unsqueeze",
"MatMul",
"Softmax",
"Transpose",
"ReduceMean",
"Reshape",
"Constant",
"Mul"
],
"weight_type": "QInt8"
},
"encoder_model": {
"op_types": [
"Div",
"Erf",
"Add",
"Shape",
"Conv",
"Sub",
"Sqrt",
"Gather",
"Pow",
"Concat",
"Reshape",
"Unsqueeze",
"MatMul",
"Softmax",
"Transpose",
"ReduceMean",
"Constant",
"Mul"
],
"weight_type": "QUInt8"
}
}
}