HAI-Lab/MLLMJailbreak-en
收藏官方服务:
资源简介:
MLLMJailbreak-en数据集是一个基于现有英语文本数据集构建的英语多模态大语言模型(MLLM)越狱攻击数据集。它使用FigStep、MML和SI-Attack方法生成,专门用于测试和评估多模态大语言模型在对抗性攻击下的安全性和鲁棒性。该数据集旨在为研究人员提供用于越狱攻击的多样化提示和图像组合,以促进模型安全性的改进。
The MLLMJailbreak-en dataset is an English-based multimodal large language model (MLLM) jailbreak attack dataset created using existing English text datasets with FigStep, MML, and SI-Attack methods. It is designed for testing and evaluating the security and robustness of MLLMs against adversarial attacks, providing diverse prompt-image combinations for jailbreak scenarios to advance model safety research.
提供机构:
HAI-Lab


