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