GuardReasoner-VLTrain
收藏arXiv2025-09-30 收录
下载链接:
https://github.com/yueliu1999/GuardReasoner-VL/
下载链接
链接失效反馈官方服务:
资源简介:
该数据集是一个包含123,000个样本和631,000个推理步骤的推理语料库,涵盖了文本、图像以及文本图像输入。为了提升样本的多样性和难度,该数据集采用了拒绝采样和数据增强的方法。规模上,该数据集包含了123,000个样本和631,000个推理步骤,其任务是基于推理的审核决策制定。
This dataset is a reasoning corpus containing 123,000 samples and 631,000 reasoning steps, covering text, images, and text-image multimodal inputs. To enhance the diversity and difficulty of the samples, rejection sampling and data augmentation strategies are adopted. With such a scale, this dataset is designed for reasoning-based audit decision-making tasks.



