A novel end-to-end trained large multimodal model that combines a vision encoder and Vicuna for general-purpose visual and language understanding.