Llama Attack

Introduction

This repo contains apps and implementations of adversarial attacks against Large Language Models and Vision Language Models.

Implementations

• Greedy Coordinate Gradient (by nanoGCG)
• Don't Say No (modified verstion of nanoGCG)
• Visual Embedding Attack (different images with similar embeddings)

Apps

Apps are created using Streamlit for implementations.

• GCG app
• DSN app
• VisEmb app

Models

• Llama 3.1
• Qwen2-VL