Reproduce “RepoST: Scalable Repository-Level Coding Environment Construction with Sandbox Testing” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yiqing Xie, Alex Xie, Divyanshu Sheth, Pengfei Liu, Daniel Fried, Carolyn Rose
Carnegie Mellon University researchers introduce REPOST, an automated framework for constructing repository-level coding environments that enables execution feedback through sandbox testing, creating the largest dataset (7,415 functions from 824 repositories) for training and evaluating code generation models while achieving up to 5.5% improvement on standard benchmarks through rejection sampling with execution feedback.
Key insights
- Sandbox testing is more scalable than integration testing for repository-level code
- Rejection sampling with execution feedback improves model performance compared to standard supervised fine-tuning
- Generated docstrings can enhance code generation performance versus original or no docstrings
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.