LLaVA-Med

LLaVA-Med is a biomedical vision-language model from Microsoft Research, released in 2023. It adapts the LLaVA recipe to medicine: a vision encoder aligned to a language model through curriculum learning on biomedical figure-caption pairs, then instruction-tuned on GPT-4-generated biomedical QA.

The model answers questions about biomedical images such as radiology scans and pathology figures, and it was presented as a step toward multimodal medical assistants that can read figures the way clinicians do.

LLaVA-Med is a research release with open weights. Like all medical AI, its outputs are not clinically validated and it should not be used for diagnosis without proper evaluation.