distilvit / quantize_config.json
tarekziade's picture
New training round from scratch
a7f7ce0 verified
raw
history blame
3.24 kB
{
"per_channel": false,
"reduce_range": false,
"per_model_config": {
"decoder_model": {
"op_types": [
"Sub",
"Concat",
"Squeeze",
"Where",
"Slice",
"Sqrt",
"Mul",
"Pow",
"Div",
"MatMul",
"Gather",
"Softmax",
"Transpose",
"ConstantOfShape",
"Reshape",
"Cast",
"Tanh",
"Unsqueeze",
"Add",
"Split",
"Shape",
"Gemm",
"Constant",
"Range",
"ReduceMean"
],
"weight_type": "QInt8"
},
"decoder_model_merged": {
"op_types": [
"Sub",
"Concat",
"Squeeze",
"Where",
"Slice",
"Sqrt",
"Mul",
"Pow",
"Div",
"If",
"MatMul",
"Gather",
"Softmax",
"Transpose",
"ConstantOfShape",
"Reshape",
"Cast",
"Tanh",
"Unsqueeze",
"Add",
"Split",
"Shape",
"Gemm",
"Constant",
"Range",
"ReduceMean"
],
"weight_type": "QInt8"
},
"decoder_with_past_model": {
"op_types": [
"Sub",
"Concat",
"Squeeze",
"Where",
"Slice",
"Sqrt",
"Mul",
"Pow",
"Div",
"MatMul",
"Gather",
"Softmax",
"Transpose",
"ConstantOfShape",
"Reshape",
"Cast",
"Tanh",
"Unsqueeze",
"Add",
"Split",
"Shape",
"Gemm",
"Constant",
"Range",
"ReduceMean"
],
"weight_type": "QInt8"
},
"encoder_model": {
"op_types": [
"Sub",
"Concat",
"Where",
"Slice",
"Sqrt",
"Mul",
"Pow",
"Div",
"MatMul",
"Gather",
"Erf",
"Softmax",
"Transpose",
"Expand",
"ConstantOfShape",
"Reshape",
"Unsqueeze",
"Conv",
"Add",
"Shape",
"Equal",
"Constant",
"ReduceMean"
],
"weight_type": "QUInt8"
}
}
}