MedChatZH is a Chinese medical conversation dataset released by East China University of Science and Technology in 2023. The related paper results are "MedChatZH: A tuning LLM for traditional Chinese medicine consultations", which aims to improve the ability to understand and generate Chinese (especially in TCM scenarios) medical consultation dialogues through continuous pre-training of TCM classics and fine-tuning of medical instruction data. The data comes from over 1,000 TCM classics and medical notes, as well as over 7 million Chinese medical instructions collected from the internet and multiple Chinese hospitals. Combined with the BELLE-3.5M general instructions, 763,629 medical instructions and 1,305,194 general instructions were screened and cleaned to form the med-mix-2M dataset for dialogue fine-tuning. Together with the TCM classics corpus, they serve the two stages of continued pre-training and instruction fine-tuning.

Citation

@article{tan2024medchatzh, title={MedChatZH: A tuning LLM for traditional Chinese medicine consultations}, author={Tan, Yang and Zhang, Zhixing and Li, Mingchen and Pan, Fei and Duan, Hao and Huang, Zijie and Deng, Hua and Yu, Zhuohang and Yang, Chen and Shen, Guoyang and others} journal={Computers in Biology and Medicine}, pages={108290}, year={2024}, publisher={Elsevier} }