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