Foreshhh/vlsbench
收藏Hugging Face2024-12-13 更新2024-12-14 收录
下载链接：
https://hf-mirror.com/datasets/Foreshhh/vlsbench
下载链接
链接失效反馈官方服务：
资源简介：
VLSBench数据集用于研究多模态安全问题，特别是评估当前多模态大语言模型（MLLMs）的安全性。数据集包含指令ID、指令、图像路径、类别、子类别和来源等特征。数据集的规模在1K到10K之间，语言为英语。数据集仅用于研究目的，不涉及用户去匿名化，因此不构成人类受试者研究。
The VLSBench dataset is used to study information leakage in multimodal safety, particularly visual leakage. The dataset includes features such as instruction ID, instruction, image path, category, sub-category, and source. The purpose of creating the dataset is to evaluate the safety of current multimodal large language models, and the data collection process does not involve user deanonymization.
提供机构：
Foreshhh
