Synthetic SFT Data

该数据集经过精心策划，旨在让大型多模态模型通过生成视觉思维、批判和改进，来进行长时间的多模态推理。该数据集包含了多种提示语和生成的图片，以支持复杂的视觉推理任务，同时采用了多种最先进的模型来进行图像生成和批判。具体任务包括带有中间视觉子目标和自我批判的图像生成。

This meticulously curated dataset is designed to enable large multimodal models to conduct long-duration multimodal reasoning by generating visual thoughts, critiques and revisions. It contains diverse prompts and generated images to support complex visual reasoning tasks, and leverages multiple state-of-the-art models for image generation and critique. Specific tasks include image generation with intermediate visual sub-goals and self-critique.