LiteLlama-460M-1T / quantize_config.json
Xenova's picture
Xenova HF staff
Upload folder using huggingface_hub
647e94b
raw
history blame
890 Bytes
{
"per_channel": false,
"reduce_range": false,
"per_model_config": {
"model": {
"op_types": [
"Shape",
"Div",
"Cast",
"MatMul",
"Sqrt",
"Expand",
"Mul",
"Less",
"Transpose",
"ConstantOfShape",
"Softmax",
"Unsqueeze",
"Squeeze",
"Reshape",
"Equal",
"Add",
"Gather",
"Pow",
"Neg",
"Sigmoid",
"Concat",
"Sub",
"Slice",
"Range",
"Constant",
"ReduceMean",
"Where"
],
"weight_type": "QInt8"
}
}
}