Reproduce “JURIST: Training Small Data-Reweighted Reasoning Judges for Test-Time Scaling in 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
Peijia Qin, Ruiyi Zhang, Qi Cao, Pengtao Xie
JURIST trains a compact judge to select passing programs from multiple code candidates. It learns which earlier training examples to emphasize by measuring their effect on a small, deployment-aligned meta set. On LiveCodeBench Hard, its 1.7B model reaches 73.8% selection accuracy, matching Claude Sonnet 4.6, at 463 times lower judging cost per evaluation pass; this excludes candidate generation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.