onnx-bloomz-560m-sft-chat / quantize_config.json
Felladrin's picture
Add model files
ee81e66
raw
history blame contribute delete
890 Bytes
{
"per_channel": false,
"reduce_range": false,
"per_model_config": {
"model": {
"op_types": [
"Where",
"Range",
"ConstantOfShape",
"Sub",
"Gather",
"Cast",
"Shape",
"Equal",
"ReduceMean",
"CumSum",
"Sqrt",
"Unsqueeze",
"Reshape",
"Softmax",
"Add",
"Mul",
"Expand",
"Less",
"Squeeze",
"Tanh",
"Div",
"Constant",
"Slice",
"MatMul",
"Transpose",
"Concat",
"Pow"
],
"weight_type": "QInt8"
}
}
}