whisper-small-sr-onnx / quantize_config.json
Sagicc's picture
Upload 19 files
efd3d9f verified
raw
history blame
2.84 kB
{
"per_channel": false,
"reduce_range": false,
"per_model_config": {
"encoder_model": {
"op_types": [
"Transpose",
"Div",
"Shape",
"Concat",
"Softmax",
"Sqrt",
"Pow",
"ReduceMean",
"Sub",
"Unsqueeze",
"Mul",
"Add",
"Conv",
"MatMul",
"Constant",
"Reshape",
"Erf",
"Gather"
],
"weight_type": "QUInt8"
},
"decoder_model_merged": {
"op_types": [
"ReduceMean",
"Unsqueeze",
"Less",
"Gather",
"Concat",
"Cast",
"Range",
"Mul",
"Equal",
"Slice",
"Expand",
"Constant",
"Erf",
"Transpose",
"Div",
"Softmax",
"Squeeze",
"Reshape",
"If",
"Shape",
"Sqrt",
"Pow",
"ConstantOfShape",
"Add",
"MatMul",
"Where",
"Sub"
],
"weight_type": "QInt8"
},
"decoder_model": {
"op_types": [
"ReduceMean",
"Unsqueeze",
"Less",
"Gather",
"Concat",
"Cast",
"Range",
"Mul",
"Equal",
"Slice",
"Expand",
"Constant",
"Erf",
"Transpose",
"Div",
"Softmax",
"Squeeze",
"Reshape",
"Shape",
"Sqrt",
"Pow",
"ConstantOfShape",
"Add",
"MatMul",
"Where",
"Sub"
],
"weight_type": "QInt8"
},
"decoder_with_past_model": {
"op_types": [
"Concat",
"Div",
"Shape",
"Transpose",
"Softmax",
"Sqrt",
"Pow",
"ReduceMean",
"Erf",
"Unsqueeze",
"Mul",
"Add",
"Slice",
"MatMul",
"Constant",
"Reshape",
"Sub",
"Gather"
],
"weight_type": "QInt8"
}
}
}