ZhongJing: A Locally Deployed Large Language Model for Traditional Chinese Medicine and Corresponding Evaluation Methodology: A Large Language Model for data fine-tuning in the field of Traditional Chinese Medicine, and a new evaluation method called TCMEval are proposed

The success of ChatGPT has showcased the potential applications of Large Language Models (LLMs) in the field of Traditional Chinese Medicine (TCM), encompassing areas such as medical diagnosis, adjunctive therapy, and TCM talent cultivation. However, the current challenges, including hardware constraints, insufficient model domain knowledge, and difficulties in domain-specific evaluation, have constrained the fusion of LLMs with TCM. In an attempt to address these issues, this paper introduces ZhongJing, a domain-specific LLM fine-tuned within the domain of TCM, capable of generating responses at a rate of 8 tokens per second, smoothly operating on local personal computers. To assess the model's domain expertise, this paper introduces the TCMEval evaluation method, designed concerning medical students' exams. Experimental results demonstrate that ZhongJing achieves a 6.49 TCMEval Score improvement over Chinese-LLaMA2 in the field of TCM, indicating the model's ability to generate more specialized responses compared to baseline models.