UCSC-VLAA/VLAA-Thinking
收藏Hugging Face2025-09-27 更新2025-04-12 收录
下载链接：
https://hf-mirror.com/datasets/UCSC-VLAA/VLAA-Thinking
下载链接
链接失效反馈官方服务：
资源简介：
VL-Thinking是一个从R1派生的视觉指令微调数据集，用于训练可思考的大型语言模型。数据集涵盖了来自不同领域（如数学、通用）和不同类型（如封闭式、开放式）的问题。为了确保更高的多样性，数据集中的所有图像都是唯一的。数据生成过程包括图像标注、视觉语言链生成、答案重写和答案验证。
VL-Thinking is a visual instruction tuning dataset derived from R1, used for training thinkable large language models. The dataset covers questions from various domains such as math and general, and different types like close-ended and open-ended questions. All images in the dataset are unique to ensure higher diversity. The data generation process includes image captioning, visual-language chain generation, answer rewriting, and answer verification.
提供机构：
UCSC-VLAA
