LLaVA is a novel end-to-end trained large multimodal model that combines a vision encoder and Vicuna, achieving impressive chat capabilities, emulating the spirit of multimodal GPT-4, and achieving new highest accuracy in scientific question answering. LLaVA's use cases include multimodal chat in daily user applications and multimodal reasoning in the scientific domain. LLaVA's data, code, and checkpoints are limited to research use and follow the licenses of CLIP, LLaMA, Vicuna, and GPT-4.