GitHub

LLaVA-CoT is a novel Vision-Language Model (VLM) designed to perform autonomous, multi-stage reasoning, enabling it to tackle complex visual question-answering tasks by independently engaging in sequential stages of summarization, visual interpretation, logical reasoning, and conclusion generation.