Reproduce “LLM4CVE: Enabling Iterative Automated Vulnerability Repair with Large Language Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Mohamad Fakih, Rahul Dharmaji, Halima Bouzidi, Gustavo Quiros Araya, Oluwatosin Ogundare, Mohammad Abdullah Al Faruque
LLM4CVE introduces an iterative, automated pipeline that utilizes Large Language Models (LLMs) to repair real-world software vulnerabilities in C code with minimal human intervention. The system, employing LoRA fine-tuning and automated feedback, significantly reduces vulnerability fix times from weeks to hours and successfully mitigates actual exploits.
Key insights
- Iterative refinement with automated feedback significantly improves LLM-generated patch quality, increasing semantic similarity to ground truth fixes.
- LoRA fine-tuning enables open-source LLMs to achieve performance comparable to or surpassing proprietary models for vulnerability repair.
- Integrating prompt engineering with an automated feedback loop and LLM augmentation creates a robust framework for autonomously generating compilable and effective security patches.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.