Clara is a Llama 3.1 based finetuned model specifically for vulnerability detection using a custom dataset based on PrimeVul

Our goal is to find a unique approach to solve the challenges of software vulnerability detection. Traditional methods often fall short, struggling with complex code and evolving threats. We aim to leverage the power of Large Language Models (LLMs) to create a system that can identify vulnerabilities that are typically missed by conventional tools.

This project builds upon several important contributions to the field.

We used unsloth for post-training, benefiting from its efficient CPU and GPU training and inference capabilities. Our Large Language Model is based on LLAMA 3.1, and its vulnerability detection capabilities were enhanced through fine-tuning with the PrimeVul dataset. We are grateful for the availability of these resources.