HPAI-BSC/medqa-cot-llama31

MedQA-CoT数据集是一个基于MedQA的合成增强响应数据集，用于提高训练分割的回答质量。该数据集通过使用Llama-3.1-70B-Instruct模型生成链式思维（CoT）答案，并创建了一个自定义提示以及一个手工制作的少量示例列表。对于多选答案，模型会重述并解释问题，然后解释每个选项与问题的关系，最后总结解释得出最终解决方案。在数据生成过程中，模型还会给出解决方案和参考答案。如果模型未能生成正确响应，则会重新生成解决方案，直到生成正确的响应。更多信息可以在相关论文中找到。

The MedQA-CoT dataset is a synthetically enhanced response dataset based on MedQA, used to improve the quality of answers from the training splits. The dataset leverages the Llama-3.1-70B-Instruct model to generate Chain of Thought (CoT) answers, creating a custom prompt along with a hand-crafted list of few-shot examples. For a multichoice answer, the model is asked to rephrase and explain the question, then explain each option in relation to the question, and finally summarize this explanation to arrive at the final solution. During the data generation process, the model is also provided with the solution and the reference answer. If the model fails to generate correct responses, the solutions are regenerated until a correct response is produced. More details can be found in the paper.