Huatuo-Llama-Med-Chinese provides instruction-tuned Chinese medical LLMs (LLaMA, Bloom, Alpaca-Chinese) with LoRA weights and inference scripts. It includes data handling, model training details, and usage examples, focusing on medical QA and knowledge-tuning.
Collecting history — the radar snapshots this repo daily. The trend line appears after 3 days of data (1 so far).
What it is
Huatuo-Llama-Med-Chinese is a repository that open-sources instruction-tuned large language models with Chinese medical knowledge, including LLaMA-, Alpaca-Chinese-, Bloom-, and Huazi-based variants. It provides datasets built from medical knowledge bases and literature, and scripts for finetuning and inference.
How it works
The project uses instruction-tuning/finetuning of base models with LoRA adapters, leveraging a Chinese medical knowledge base and medical literature. It combines knowledge retrieval prompts and structured data to produce more trustworthy medical responses.
Getting started
First steps shown:
pip install -r requirements.txt
It lists base models and LoRA weight download sources for Huatuo/BenTsao, Bloom, Alpaca-Chinese, and LLaMA variants. It also provides scripts for inference:
bash ./scripts/infer.sh
bash ./scripts/infer-literature-single.sh
bash ./scripts/infer-literature-multi.sh
Run inference via infer.py with base_model, lora_weights, instruct_dir, and prompt_template parameters as shown in the README excerpt.
Recent releases
Latest releases section shows: "LATEST 0: - none".
Traction
Stars: 4983. forks: 498. open_issues: 28. created: 2023-03-31. last_push: 2026-07-04.
Behind the repo
No linked startup/company information provided in the facts block.
Caveats
License: Apache-2.0. Language: Python. Requires Python 3.9+. It mentions training on A100-80GB hardware, 10 training rounds, roughly 2h17m total, and about 40G VRAM at batch_size 128. It notes that LoRA weights are distributed via Baidu Drive and Hugging Face links, and that data quality may have issues and will be iterated.