RLEP: Reinforcement Learning with Experience Replay for LLM Reasoning

TL;DR

RLEP introduces a two-phase reinforcement learning framework for large language models that replays verified successful trajectories to improve training efficiency and reasoning accuracy, achieving state-of-the-art results on math benchmarks.

Contribution

The paper proposes RLEP, a novel reinforcement learning method that combines experience replay with verified trajectories to enhance LLM reasoning performance.

Findings

Faster convergence in training.

Improved accuracy on math benchmarks.

Effective replay of high-quality reasoning paths.

Abstract

Reinforcement learning (RL) for large language models is an energy-intensive endeavor: training can be unstable, and the policy may gradually drift away from its pretrained weights. We present \emph{RLEP}\, -- \,Reinforcement Learning with Experience rePlay\, -- \,a two-phase framework that first collects verified trajectories and then replays them during subsequent training. At every update step, the policy is optimized on mini-batches that blend newly generated rollouts with these replayed successes. By replaying high-quality examples, RLEP steers the model away from fruitless exploration, focuses learning on promising reasoning paths, and delivers both faster convergence and stronger final performance. On the Qwen2.5-Math-7B base model, RLEP reaches baseline peak accuracy with substantially fewer updates and ultimately surpasses it, improving accuracy on AIME-2024 from 38.2% to 39.9%, on AIME-2025 from 19.8% to 22.3%, and on AMC-2023 from 77.0% to 82.2%. Our code, datasets, and checkpoints are publicly available at https://github.com/Kwai-Klear/RLEP to facilitate reproducibility and further research.