Improving RL Exploration for LLM Reasoning through Retrospective Replay
This addresses a bottleneck in RL for LLM post-training, particularly for complex reasoning tasks, with incremental improvements to existing methods.
The paper tackles the problem of ineffective exploration in reinforcement learning for large language models, where early promising solution ideas get suppressed and aren't revisited later. The proposed Retrospective Replay-based Reinforcement Learning (RRL) algorithm improves exploration efficiency, significantly enhancing RL effectiveness for complex reasoning tasks like mathematical reasoning and code generation, and also improves RLHF performance for safety and helpfulness.
Reinforcement learning (RL) has increasingly become a pivotal technique in the post-training of large language models (LLMs). The effective exploration of the output space is essential for the success of RL. We observe that for complex problems, during the early stages of training, the model exhibits strong exploratory capabilities and can identify promising solution ideas. However, its limited capability at this stage prevents it from successfully solving these problems. The early suppression of these potentially valuable solution ideas by the policy gradient hinders the model's ability to revisit and re-explore these ideas later. Consequently, although the LLM's capabilities improve in the later stages of training, it still struggles to effectively address these complex problems. To address this exploration issue, we propose a novel algorithm named Retrospective Replay-based Reinforcement Learning (RRL), which introduces a dynamic replay mechanism throughout the training process. RRL enables the model to revisit promising states identified in the early stages, thereby improving its efficiency and effectiveness in exploration. To evaluate the effectiveness of RRL, we conduct extensive experiments on complex reasoning tasks, including mathematical reasoning and code generation, and general dialogue tasks. The results indicate that RRL maintains high exploration efficiency throughout the training period, significantly enhancing the effectiveness of RL in optimizing LLMs for complicated reasoning tasks. Moreover, it also improves the performance of RLHF, making the model both safer and more helpful.