Reproduce “PATCH: Empowering Large Language Model with Programmer-Intent Guidance and Collaborative-Behavior Simulation for Automatic Bug Fixing” 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
Yuwei Zhang, Zhi Jin, Ying Xing, Ge Li, Fang Liu, Jiaxin Zhu, Wensheng Dou, Jun Wei
PATCH turns Java bug fixing from one-shot generation into a staged workflow that supplies repository context and programmer intent, then uses prompted tester, developer, and reviewer agents to report, diagnose, generate, and revise patches. On the BFP benchmark, gpt-3.5-turbo-0125 reaches 33.97% Fix@1, versus GPT-4’s 19.96%; this exact-match result covers single-hunk bugs with fault locations given.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.