Llama3, Codestral, and DeepSeek-R1 Performance in Cross-Domain Vulnerability Detection

This report synthesises findings from 4 peer-reviewed papers addressing the following research question: How do Llama3, Codestral, and Deepseek R1 perform on cross-domain vulnerability detection (e.g., C/C++ vs. Python) when fine-tuned on adversarially perturbed datasets, as measured by domain-specific. 11 claims were extracted from source literature; 11 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 8.2/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: How do Llama3, Codestral, and Deepseek R1 perform on cross-domain vulnerability detection (e.g., C/C++ vs. Python) when fine-tuned on adversarially perturbed datasets, as measured by domain-specific accuracy and F1 score?

Autonomous literature synthesis. Automated review score: 8.2/10. Full text and citation available at Assignee Research.