arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2510.04072cs.LGcs.AIcs.CLstat.ML
收藏

慢-快策略优化:在更新前重新定位用于大语言模型推理

Slow-Fast Policy Optimization: Reposition-Before-Update for LLM Reasoning

Georgia Institute of Technology(佐治亚理工学院);University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);University of Manchester(曼彻斯特大学);NVIDIA;Independent(独立);Emory University(埃默里大学)

Ziyan Wang, Zheng Wang, Xingwei Qu, Qi Cheng, Jie Fu, Shengpu Tang, Minjia Zhang, Xiaoming Huo

AI总结:

本文提出慢-快策略优化框架,通过分解步骤为三个阶段,解决RL在早期训练中的稳定性与效率问题,实验表明其在推理任务中提升稳定性、减少rollouts数量并加速收敛。

摘要:

强化学习(RL)已成为增强大语言模型(LLMs)推理能力的核心。然而,如组相对策略优化(GRPO)等在线算法在早期训练中常面临问题:低质量rollouts导致的噪声梯度造成更新不稳定和探索低效。本文引入慢-快策略优化(SFPO),通过将每一步分解为三个阶段:同一批次上的短快轨迹、重新定位机制以控制离线漂移,以及最终的慢修正。这种在更新前重新定位的设计保持了目标和rollout过程不变,使SFPO与现有策略梯度流程兼容。大量实验表明,SFPO在推理RL训练中提升了稳定性,减少了rollouts数量并加速了收敛。具体而言,在数学推理基准上,SFPO比GRPO平均高出2.80分。此外,它实现了多达4.93倍的rollouts减少和4.19倍的墙钟时间减少以达到GRPO的最佳精度。项目网站可在https://slow-fast-po.github.io/上找到。

英文摘要:

Reinforcement learning (RL) has become central to enhancing reasoning in large language models (LLMs). Yet on-policy algorithms such as Group Relative Policy Optimization (GRPO) often suffer in early training: noisy gradients from low-quality rollouts lead to unstable updates and inefficient exploration. We introduce Slow-Fast Policy Optimization (SFPO), a simple yet efficient framework to address the above limitations via decomposing each step into three stages: a short fast trajectory of inner steps on the same batch, a reposition mechanism to control off-policy drift, and a final slow correction. This reposition-before-update design preserves the objective and rollout process unchanged, making SFPO plug-compatible with existing policy-gradient pipelines. Extensive experiments demonstrate that SFPO consistently improves stability, reduces number of rollouts, and accelerates convergence of reasoning RL training. Specifically, it outperforms GRPO by up to 2.80 points in average on math reasoning benchmarks. It also achieves up to 4.93\texttimes{} fewer rollouts and an up to 4.19\texttimes{} reduction in wall-clock time to match GRPO's best accuracy. Project website is available at https://slow-fast-po.github.io/.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑