%% generate tags start %% #ai %% generate tags end %% #ai/model LLaVA represents a novel end-to-end trained large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding, achieving impressive chat capabilities mimicking spirits of the multimodal GPT-4 and setting a new state-of-the-art accuracy on Science QA. [LLaVA (llava-vl.github.io)](https://llava-vl.github.io/)