XuanHuGPT: parameter-efficient fine-tuning of large language model in the field of traditional Chinese medicine

Xuming Tong ^1,2, Xiaozheng Ding ², Huiru Jia ²

⁰Faculty of Applied Sciences, Macao Polytechnic University, Macao, Macao S.A.R., China.

Chinese Medicine +

|

November 26, 2025

View abstract on PubMed

Summary

Researchers developed XuanHuGPT, a specialized Large Language Model (LLM) for Traditional Chinese Medicine (TCM). This AI model, trained on the new XhTCM dataset, shows improved accuracy and safety in TCM knowledge tasks.

Area Of Science

Artificial Intelligence
Computational Linguistics
Traditional Chinese Medicine

Background

Large Language Models (LLMs) show promise in Traditional Chinese Medicine (TCM) but lack domain-specific optimization, sufficient data, and effective fine-tuning.
Existing LLMs struggle with TCM tasks due to limited pre-training on specialized knowledge and data scarcity.

Purpose Of The Study

To develop a domain-specific LLM for Traditional Chinese Medicine (TCM) knowledge.
To create a comprehensive dataset (XhTCM) for TCM-specific AI model training.
To establish a robust evaluation framework for assessing TCM LLMs.

Main Methods

Constructed the XhTCM dataset by integrating data from ShenNong_TCM_Dataset, TCMBank, and TCMIP v2.0, comprising 100,000 structured entries.
Developed XuanHuGPT, a domain-specific LLM for TCM using Parameter-Efficient Fine-Tuning (PEFT) on the XhTCM dataset.
Implemented a comprehensive evaluation framework combining quantitative metrics (BLEU, ROUGE, METEOR, BERTScore, Embedding Distance) and qualitative expert assessments.

Main Results

XuanHuGPT demonstrated superior performance compared to general-purpose LLMs and existing TCM-specific models.
The model achieved significant improvements in accuracy, coverage, fluency, consistency, sensitivity, and safety for TCM Q&A and inference tasks.
PEFT techniques effectively balanced model performance with training costs.

Conclusions

XuanHuGPT represents a significant advancement in applying AI to Traditional Chinese Medicine.
The study provides a reproducible methodology for building intelligent TCM Q&A systems.
This work contributes to the digital transformation and global dissemination of TCM knowledge.

Keywords:

Large Language Models LoRA Parameter-efficient fine-tuning Traditional Chinese medicine XuanHuGPT