Reproduce “Enhancing Repository-Level Code Completion with Reinforcement Learning in Real-World Scenarios” 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

Zhenlong Dai, Zhuoluo Zhao, Xiu Tang, Sai Wu, Chang Yao, Zhipeng Gao, Jingyuan Chen

IntentCoder trains code models to use in-file and cross-file context for repository-level completion. It scores sampled completions against reference code and identifier overlap, then selectively updates uncertain, higher-reward tokens. On RepoBench-Python, its Qwen2.5-Coder-7B model reaches 35.93% Code Exact Match, versus 23.20% for aiXcoder-7B-v2; this measures exact line matching, not functional correctness.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.