Reproduce “GARL: Efficient Dual-Agent Mathematical Reasoning via Guidance-Aware Reinforcement Learning” 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
Yuxiang Feng, Zheng Wang, Bo Xu
The authors introduce GARL, which pairs an Actor that solves a problem with a Coach that checks its solution and provides error diagnosis, a plan, and cautions for another attempt. The Actor and Coach are trained through reinforcement learning and preference optimization. On five benchmarks, Qwen3-4B averages 72.0% pass@1 versus 69.9% for GRPO, within this evaluation.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.