CMed-Baichuan: Task Explanation-Enhanced Prompt Method on PromptCBLUE Benchmark

Large Language Models (LLMs) have received widespread attention from academia and industry for their excellent performance on NLP tasks. Due to the knowledge-intensive nature of the medical field, previous studies proposed various fine-tuning methods and fine-tuned domain LLMs to align the general LLMs into specific domains. However, they ignored the difficulty of understanding the medical task requirements, that is LLMs are expected to give answers in the situation of not fully understanding the requirements of the task itself and instructions. So, in this paper, we argue that the explanation of task requirements is important to improve LLM’s understanding. Moreover, we proposed a task explanation-enhanced prompt method and introduced a medical LLM, CMed-Baichuan. In addition, we evaluated our model on the PromptCBLUE benchmark, which is the first LLM evaluation benchmark covering 18 Chinese medical NLP tasks, and the experimental results show that our model achieved state-of-the-art performance on overall score, and also demonstrate the importance of task explanation. Our code is publicly available at https://github.com/naginoa/CMed-Baichuan .