MING

Why we included this project

MING is a Chinese-language medical consultation model that has been instruction-tuned on top of open base models like Qwen and Bloomz rather than built from scratch. It handles two kinds of work: answering medical questions and analyzing cases, and running multi-turn consultations where it keeps asking follow-up questions until it offers a diagnosis and advice. Because the weights come in several sizes, from a 1.8B version up to a mixture-of-experts model, a small team can pick one that fits its compute budget instead of renting a cluster. The repo also ships the training code, so it doubles as a working example of fine-tuning an open base model for a narrow, high-stakes domain. That makes it useful both as a deployable assistant for Chinese health products and as a reference for teams planning their own domain-specific fine-tuning.