The article presents the results of fine-tuning large language models (LLMs) to detect vulnerabilities in source code. The authors used WizardCoder, an improvement of the state-of-the-art LLM StarCoder, and adapted it for vulnerability detection. They increased the training speed of the model without impacting its performance, optimized the training procedure, and improved performance on difficult vulnerability detection datasets. The study demonstrates the potential of transfer learning by fine-tuning large pre-trained language models for specialized tasks like source code analysis.
Publication date: 30 Jan 2024
Project Page: https://arxiv.org/abs/2401.17010v1
Paper: https://arxiv.org/pdf/2401.17010
