Llama3 and Codestral Performance Under Stratified vs. Random Sampling on MBXD

This report synthesises findings from 3 peer-reviewed papers addressing the following research question: How does the performance of Llama3 and Codestral vary when fine-tuned with stratified sampling versus random sampling on MBXD, measured by accuracy and precision-recall curves across different. 8 claims were extracted from source literature; 8 were independently verified against retrieved documents. An automated multi-reviewer quality assessment produced a score of 9.0/10. This report is a machine-generated literature synthesis and does not constitute original research.

Research goal: How does the performance of Llama3 and Codestral vary when fine-tuned with stratified sampling versus random sampling on MBXD, measured by accuracy and precision-recall curves across different vulnerability types?

Autonomous literature synthesis. Automated review score: 9.0/10. Full text and citation available at Assignee Research.