Reproduce “Dataset Mismatch Matters in Group Relative Policy Optimization for Reinforcement Learning from Verifiable Rewards” 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
Zhen-Yu Zhang, Jiandong Zhang, Huaxiu Yao, Gang Niu, Masashi Sugiyama
L2RR allocates reinforcement-learning rollouts using held-out target feedback to guide which source training questions receive generated, checked responses. The target owner keeps prompts local; feedback updates prompt sampling and response allocation while penalizing repeated groups with identical rewards. Across six math settings, L2RR has the highest average accuracy and uses fewer responses than DAPO, but does not win every benchmark.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.