gpt2-magic-card-web / quantize_config.json
gstaff's picture
Upload files for transformers.js
28ea93b
raw
history blame
2.4 kB
{
"per_channel": true,
"reduce_range": true,
"per_model_config": {
"decoder_model": {
"op_types": [
"Shape",
"Pow",
"Concat",
"Div",
"Add",
"Cast",
"Slice",
"Sub",
"Gather",
"ConstantOfShape",
"MatMul",
"Reshape",
"Where",
"Unsqueeze",
"Tanh",
"Transpose",
"Squeeze",
"Softmax",
"ReduceMean",
"Constant",
"Mul",
"Split",
"Sqrt",
"Range",
"Gemm"
],
"weight_type": "QInt8"
},
"decoder_model_merged": {
"op_types": [
"Shape",
"Pow",
"If",
"Concat",
"Div",
"Add",
"Cast",
"Slice",
"Sub",
"Gather",
"ConstantOfShape",
"MatMul",
"Reshape",
"Where",
"Unsqueeze",
"Tanh",
"Transpose",
"Squeeze",
"Softmax",
"ReduceMean",
"Constant",
"Mul",
"Split",
"Sqrt",
"Range",
"Gemm"
],
"weight_type": "QInt8"
},
"decoder_with_past_model": {
"op_types": [
"Shape",
"Pow",
"Concat",
"Div",
"Add",
"Cast",
"Slice",
"Sub",
"Gather",
"ConstantOfShape",
"MatMul",
"Reshape",
"Where",
"Unsqueeze",
"Tanh",
"Transpose",
"Squeeze",
"Softmax",
"ReduceMean",
"Constant",
"Mul",
"Split",
"Sqrt",
"Range",
"Gemm"
],
"weight_type": "QInt8"
}
}
}