{ "per_channel": false, "reduce_range": false, "per_model_config": { "encoder_model": { "op_types": [ "Mul", "Pow", "Add", "Div", "Softmax", "Conv", "Erf", "Sub", "Transpose", "Reshape", "Gather", "ReduceMean", "Sqrt", "Constant", "MatMul", "Concat", "Unsqueeze", "Shape" ], "weight_type": "QUInt8" }, "decoder_model": { "op_types": [ "Pow", "Expand", "Transpose", "Less", "Softmax", "Range", "ConstantOfShape", "Mul", "Add", "Sub", "Where", "Gather", "Squeeze", "MatMul", "Erf", "Cast", "Reshape", "Concat", "Unsqueeze", "Div", "Slice", "Equal", "ReduceMean", "Sqrt", "Constant", "Shape" ], "weight_type": "QInt8" }, "decoder_with_past_model": { "op_types": [ "Mul", "Pow", "Add", "Div", "Softmax", "Slice", "Sub", "Erf", "Transpose", "Reshape", "Gather", "ReduceMean", "Sqrt", "Concat", "Constant", "MatMul", "Unsqueeze", "Shape" ], "weight_type": "QInt8" }, "decoder_model_merged": { "op_types": [ "Pow", "Expand", "Transpose", "Less", "Softmax", "Range", "ConstantOfShape", "Mul", "Add", "Sub", "Where", "Gather", "Squeeze", "MatMul", "Erf", "Cast", "Reshape", "Concat", "Unsqueeze", "Div", "If", "Slice", "Equal", "ReduceMean", "Sqrt", "Constant", "Shape" ], "weight_type": "QInt8" } } }