Reproduce “Small Agents with Semantic Search: Efficient Multilingual Code Localization” 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
Maxence Lasbordes, Aarush Sinha, Raphael Sourty, Amélie Chatelain, Djamé Seddah
The authors evaluate ColGREP, a local semantic code-search tool used by a compact agent to find files from natural-language descriptions, inspect candidates, and submit paths for bug localization. On SWE-bench Lite, Qwen3-1.7B with reinforcement learning scores 67.29 file F1, versus 63.75 with grep. This measures localization against patch-changed files, not successful repair.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.