Reproduce “HELO-APR: Enhancing Low-Resource Program Repair through Cross-Lingual Knowledge Transfer” 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
Zhipeng Wang, Boyang Yang, Yidong Wan, Liuye Guo, You Lv, Tao Zheng, Zhuowei Wang, Tieke He
HELO-APR transfers repair behavior from C++ examples to Ruby and Rust by constructing verified target-language bug-fix pairs and training models in three stages, including explicit cross-language alignment. On xCodeEval, CodeLlama-7B-Instruct achieves average Pass@1 of 11.97% versus 5.12% for the strongest baseline. Its average target-language compilation rate reaches 91.98%, up from 49.77% zero-shot; these results cover C++-to-Ruby/Rust transfer on tested 7B models.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.