Visual-Adversarial-Examples-Jailbreak-Large-Language-Models
Repository for visual adversarial examples that jailbreak large language models
GraphCanon updated 2w · GitHub synced 2w
Decision brief
This tool focuses on generating and studying visual adversarial examples designed to exploit vulnerabilities in large language models such as MiniGPT-4.
Good fit when
- If your goal is to research the resilience of Large Language Models against visual inputs that could mislead or 'jailbreak' their usual behavior.
- For scholars and researchers studying model security and robustness, particularly focusing on how image data can affect language models.
Avoid when
- Avoid using this tool if your objective does not involve security testing or research concerning visual-linguistic interactions with large language models.
- If you are looking to enhance general-purpose AI applications without considering adversarial attacks, other frameworks would be more appropriate.
Observed Jul 16, 2026 · Source: enrich:decision_facts
Verify the decision
Maintenance and security
Full trust report- Maintenance
- Dormant (813d since push)
- As of 2w
- Provenance
- Not a fork · Personal account
- As of 2w
- Security (OSV)
- No lockfile
- As of 1mo
Public GitHub metadata and optional OSV scans. Signals, not a guarantee. Trust methodology.
Install
pip install Visual-Adversarial-Examples-Jailbreak-Large-Language-ModelsSimilar tools
Same-category neighbours. No typed graph edges are catalogued for this tool yet.
Evidence and technical details
Sourced facts, taxonomy, compatibility claims, README excerpt, and machine-readable endpoints.
Overview
This repository contains work related to creating and analyzing visual adversarial examples aimed at disrupting the functioning of large language models, specifically showcasing attacks on MiniGPT-4.
Capability facts
- Languages
- python
Source: github.language · Aug 5, 2026
Categories
Tags
README
Installation
We take MiniGPT-4 (13B) as the sandbox to showcase our attacks. The following installation instructions are adapted from the MiniGPT-4 repository.
1. Set up the environment
git clone https://github.com/Unispac/Visual-Adversarial-Examples-Jailbreak-Large-Language-Models.git
cd Visual-Adversarial-Examples-Jailbreak-Large-Language-Models
conda env create -f environment.yml
conda activate minigpt4
2. Prepare the pretrained weights for MiniGPT-4
As we directly inherit the MiniGPT-4 code base, the guide from the MiniGPT-4 repository can also be directly used to get all the weights.
-
Get Vicuna: MiniGPT-4 (13B) is built on the v0 version of Vicuna-13B. Please refer to this guide from the MiniGPT-4 repository to get the weights of Vicuna.
Then, set the path to the vicuna weight in the model config file here at Line 16.
-
Get MiniGPT-4 (the 13B version) checkpoint: download from here.
Then, set the path to the pretrained checkpoint in the evaluation config file in eval_configs/minigpt4_eval.yaml at Line 11.
For agents
This page has a .md twin and JSON over the API.