Your work is excellent, but I have a question. When calculating accuracy, the value of "model_answer" is not included in the response of the model. How should we handle this situation.