XrayGLM

Why we included this project

XrayGLM is one of the few openly available multimodal models tuned for chest X-ray reading in Chinese. It starts from VisualGLM-6B and fine-tunes it on X-ray images paired with diagnostic radiology reports, so given a chest radiograph it writes a natural-language summary of the findings rather than a class label. The repo is worth more than the weights alone: it also ships cleaned bilingual training data built from MIMIC-CXR and OpenI, along with scripts to reproduce the LoRA fine-tuning on your own data. That makes it a solid starting point for a radiology-assist prototype and a quick way to see how a 6B-scale medical vision-language model behaves before committing to something larger. One caveat: the CC BY-NC-SA license rules out commercial use, so review it before building on top of XrayGLM.