We fine-tuned an LLM to triage and fix insecure code

Corgea's latest blog post discusses their fine-tuned Language Learning Model (LLM), Llama 3.1 8B, aimed at enhancing enterprise application security by improving precision and privacy. The LLM is tailored for tasks like false positive detection and automated fixing, offering data isolation and avoiding reliance on third-party LLMs. A diverse dataset, innovative unsupervised training, and efficient deployment strategy enable Corgea's LLM to deliver better performance with low-cost deployment.

• Llama 3.1 8B is chosen for its size and performance.
• Fine-tuning reduces false positives by 30%.
• Updates accelerate remediation by ~80%.
• Deployment requires only a single A10 24GB GPU.
• The LLM outperforms OpenAI models by 7%.