Reproduce “MORepair: Teaching LLMs to Repair Code via Multi-Objective Fine-tuning” 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
Boyang Yang, Haoye Tian, Jiadong Ren, Hongyu Zhang, Jacques Klein, Tegawendé F. Bissyandé, Claire Le Goues, Shunfu Jin
MORepair trains code-repair models to generate corrected code alongside natural-language explanations of repair logic. It combines code-only and code-plus-guidance training objectives, then evaluates candidate patches against tests. On the constructed function-level EvalRepair-C++ benchmark, CodeLlama-13B achieves 69.5% Top-10 repair, versus 64.0% with standard fine-tuning and 58.5% untuned; this measures tested success among ten candidates, not speed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.