kdst/ELITE
收藏Hugging Face2025-09-02 更新2025-11-01 收录
下载链接:
https://hf-mirror.com/datasets/kdst/ELITE
下载链接
链接失效反馈官方服务:
资源简介:
ELITE是一个用于视觉语言模型安全性评估的全面评估基准,旨在通过使用基于量表的ELITE评估器来评估模型在多模态有害提示下的安全对齐情况。数据集包含4,587个图像-文本对,分为11个危害类别,并按照特定的评分公式进行评分。
ELITE is a comprehensive safety evaluation benchmark for Vision-Language Models (VLMs) designed to evaluate the safety alignment of models under multimodal harmful prompts using the ELITE evaluator, a rubric-based automated evaluation framework. The dataset includes 4,587 image-text pairs across 11 hazard categories and is scored according to a specific scoring formula.
提供机构:
kdst



