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