metaclip-b16-fullcc2.5b / quantize_config.json
benny1923
upload model
daacb81
{
"per_channel": true,
"reduce_range": true,
"per_model_config": {
"vision_model": {
"op_types": [
"Add",
"Slice",
"Sub",
"Where",
"Shape",
"Reshape",
"MatMul",
"ConstantOfShape",
"Concat",
"Sqrt",
"Pow",
"Gather",
"Softmax",
"Div",
"Constant",
"Equal",
"Conv",
"Expand",
"Unsqueeze",
"Mul",
"Transpose",
"Cast",
"Sigmoid",
"ReduceMean"
],
"weight_type": "QUInt8"
},
"text_model": {
"op_types": [
"Add",
"Slice",
"ArgMax",
"Sub",
"Where",
"Shape",
"Squeeze",
"Reshape",
"Less",
"MatMul",
"ConstantOfShape",
"Concat",
"Sqrt",
"Range",
"Pow",
"Gather",
"Softmax",
"Flatten",
"Div",
"Constant",
"Equal",
"Expand",
"Unsqueeze",
"Mul",
"Transpose",
"Cast",
"Sigmoid",
"ReduceMean"
],
"weight_type": "QInt8"
},
"model": {
"op_types": [
"Add",
"Slice",
"ArgMax",
"Sub",
"Where",
"Shape",
"Squeeze",
"Reshape",
"Less",
"MatMul",
"ConstantOfShape",
"Concat",
"Sqrt",
"Range",
"Pow",
"Gather",
"Exp",
"Softmax",
"Flatten",
"Div",
"ReduceSum",
"Constant",
"Equal",
"Conv",
"Expand",
"Unsqueeze",
"Mul",
"Transpose",
"Cast",
"Sigmoid",
"ReduceMean"
],
"weight_type": "QUInt8"
}
}
}