{ "per_channel": false, "reduce_range": false, "per_model_config": { "decoder_model": { "op_types": [ "Add", "Erf", "Mul", "Constant", "Sqrt", "Gemm", "Split", "Where", "Transpose", "Sub", "Gather", "ConstantOfShape", "Slice", "Pow", "ReduceMean", "Concat", "Cast", "Squeeze", "Shape", "Range", "MatMul", "Reshape", "Unsqueeze", "Softmax", "Div" ], "weight_type": "QInt8" }, "decoder_model_merged": { "op_types": [ "Add", "Erf", "Mul", "Constant", "Sqrt", "Gemm", "Split", "Where", "Transpose", "Sub", "Gather", "ConstantOfShape", "Slice", "Pow", "ReduceMean", "Concat", "Cast", "Squeeze", "Shape", "Range", "MatMul", "If", "Reshape", "Unsqueeze", "Softmax", "Div" ], "weight_type": "QInt8" }, "decoder_with_past_model": { "op_types": [ "Add", "Erf", "Mul", "Constant", "Sqrt", "Gemm", "Split", "Where", "Transpose", "Sub", "Gather", "ConstantOfShape", "Slice", "Pow", "ReduceMean", "Concat", "Cast", "Squeeze", "Shape", "Range", "MatMul", "Reshape", "Unsqueeze", "Softmax", "Div" ], "weight_type": "QInt8" } } }