Reproduce “EVOR: Evolving Retrieval for Code Generation” 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
Hongjin Su, Shuyang Jiang, Yuhang Lai, Haoyuan Wu, Boao Shi, Che Liu, Qian Liu, Tao Yu
EVOR adaptively retrieves knowledge for code generation by using execution feedback to revise search queries and update its pool of retrievable examples. It tests simulated library changes and less-common languages, using pass@1—execution accuracy on one generated program. For CodeLlama, average accuracy across four benchmarks rises from 16.0% with DocPrompting to 32.2% with EVOR, limited to these designed tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.