BenCao

BenCao (originally released as HuaTuo) is a family of open-source large language models instruction-tuned with Chinese medical knowledge, published by the Research Center for Social Computing and Information Retrieval (SCIR) at Harbin Institute of Technology. The GitHub repository has more than 5,000 stars and collects model releases, training recipes and evaluation material for Chinese medical question answering.

Overview

General-purpose LLMs often stumble on Chinese medical terminology and clinical knowledge. The BenCao project addresses that by instruction-tuning strong open base models, including LLaMA, Chinese Alpaca and BLOOM variants, on Chinese medical knowledge bases and medical literature. The result is a set of downloadable models that answer medical questions in Chinese far more reliably than their bases, together with the data construction and tuning details needed to reproduce or extend them. The project grew out of SCIR’s earlier medical dialogue research and is one of the most-cited open efforts in Chinese medical NLP, frequently used as a baseline in subsequent medical LLM papers.

Key Features

• Multiple released checkpoints built on LLaMA, Chinese Alpaca, BLOOM and Huozi bases
• Instruction tuning driven by structured Chinese medical knowledge and literature
• Documented data construction pipeline for building medical instruction datasets
• Evaluation examples showing medical QA behavior in Chinese
• Open weights and recipes that researchers can fine-tune further for specific clinical domains
• Backed by an established academic NLP research group

💳 Pricing

Free and open source under the Apache 2.0 license. Models and code are published for research; anyone deploying them in a medical product remains responsible for clinical validation and regulatory compliance.

Who Is It For

Medical AI researchers and developers building Chinese-language health assistants, medical QA systems or clinical NLP prototypes who need a domain-adapted open model as a starting point.