https://github.com/VILA-Lab/M-Attack

A Simple Baseline Achieving Over 90% Success Rate Against the Strong Black-box Models of GPT-4.5/4o/o1. Paper at: https://arxiv.org/abs/2503.10635
https://github.com/VILA-Lab/M-Attack

adversarial-attack attack lvlms mllms

Last synced: about 1 month ago
JSON representation

A Simple Baseline Achieving Over 90% Success Rate Against the Strong Black-box Models of GPT-4.5/4o/o1. Paper at: https://arxiv.org/abs/2503.10635

• Host: GitHub
• URL: https://github.com/VILA-Lab/M-Attack
• Owner: VILA-Lab
• License: mit
• Created: 2025-03-11T10:50:08.000Z (3 months ago)
• Default Branch: main
• Last Pushed: 2025-04-16T09:31:07.000Z (about 2 months ago)
• Last Synced: 2025-04-16T12:31:09.223Z (about 2 months ago)
• Topics: adversarial-attack, attack, lvlms, mllms
• Language: Python
• Homepage: https://vila-lab.github.io/M-Attack-Website/
• Size: 32.4 MB
• Stars: 52
• Watchers: 0
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• StarryDivineSky - VILA-Lab/M-Attack - Attack是一个针对大型语言模型（LLM）的简单但有效的黑盒对抗攻击基线方法，它能够在GPT-4.5/4o/o1等强大的黑盒模型上实现超过90%的攻击成功率。该项目旨在提供一个易于理解和实现的对抗攻击框架，用于评估和提高LLM的鲁棒性。M-Attack的工作原理基于某种策略，能够生成对抗性输入，诱导LLM产生错误或不期望的输出。该项目提供代码和相关资源，方便研究人员复现结果并进行进一步研究。具体细节和实验结果可以在论文https://arxiv.org/abs/2503.10635中找到。M-Attack的成功表明，即使是最先进的LLM也容易受到精心设计的对抗攻击的影响。 (A01_文本生成_文本对话 / 大语言对话模型及数据)