MDIT-Dataset
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/nuo1nuo/MDIT-Bench
下载链接
链接失效反馈官方服务：
资源简介：
该数据集旨在评估大型多模态模型（LMMs）中的双重隐含毒性，包含了31万7千6百38个问题，这些问题覆盖了不同的类别和难度级别。数据集分为三个难度等级，旨在衡量模型对双重隐含毒性的敏感性。其规模宏大，涵盖了12个类别、23个子类别以及780个主题，任务是对大型多模态模型中的双重隐含毒性检测进行评估。
This dataset is designed to evaluate dual implicit toxicity in large multimodal models (LMMs). It comprises 317,638 questions covering diverse categories and difficulty levels, and is divided into three difficulty tiers to measure the model's sensitivity to dual implicit toxicity. With a substantial scale, this dataset includes 12 main categories, 23 subcategories, and 780 distinct topics, and is specifically tailored for assessing dual implicit toxicity detection in large multimodal models.
提供机构：
MDIT-Bench
