README.md

---
library_name: transformers
license: other
base_model: meta-llama/Meta-Llama-3-8B-Instruct
tags:
- llama-factory
- full
- generated_from_trainer
model-index:
- name: sft
  results: []
---

<!-- This model card has been generated automatically according to the information the Trainer had access to. You
should probably proofread and complete it, then remove this comment. -->

# REDCODER: Automated Multi-Turn Red Teaming for Code LLMs

> 🔬 A model fine-tuned for adversarial multi-turn prompt generation to induce vulnerabilities in Code LLMs.  
> 📄 [[arXiv:2507.22063](https://arxiv.org/pdf/2507.22063)] • 🧠 
> 💻 Full code & data: [GitHub – luka-group/RedCoder](https://github.com/luka-group/RedCoder)

---


## 🧠 Model Summary

**REDCODER** is a red-teaming LLM trained to engage target Code LLMs in multi-turn conversations that gradually steer them into generating **CWE vulnerabilities** (e.g., Such as path traversal, SQL injection, etc.).

This model is designed to support:
- ⚔️ **Red-teaming evaluations** for Code LLMs
- 🧪 **Security benchmarking** of model guardrails and filters
- 🧩 **Multi-turn adversarial prompt generation** in research settings

> ⚠️ This model should not be used to generate real-world exploits. Its intended use is for research, safety evaluation, and secure LLM development.

---


If you find this work useful, please cite:

```bibtex
@article{mo2025redcoder,
  title   = {REDCODER: Automated Multi-Turn Red Teaming for Code LLMs},
  author  = {Wenjie Jacky Mo and Qin Liu and Xiaofei Wen and Dongwon Jung and
             Hadi Askari and Wenxuan Zhou and Zhe Zhao and Muhao Chen},
  journal = {arXiv preprint arXiv:2507.22063},
  year    = {2025}
}
```