Entropy Ratio Clipping as a Soft Global Constraint for Stable Reinforcement Learning

TL;DR

This paper introduces Entropy Ratio Clipping (ERC), a novel global constraint mechanism for reinforcement learning that stabilizes policy updates by regulating the relative change in policy entropy, improving performance across benchmarks.

Contribution

The paper proposes ERC, a new entropy ratio-based clipping method that addresses global distributional shifts in RL, enhancing stability and performance of existing algorithms.

Findings

ERC improves stability in RL training.

ERC enhances performance across multiple benchmarks.

ERC effectively regulates policy exploration changes.

Abstract

Large language model post-training relies on reinforcement learning to improve model capability and alignment quality. However, the off-policy training paradigm introduces distribution shift, which often pushes the policy beyond the trust region, leading to training instabilities manifested as fluctuations in policy entropy and unstable gradients. Although PPO-Clip mitigates this issue through importance clipping, it still overlooks the global distributional shift of actions. To address these challenges, we propose using the entropy ratio between the current and previous policies as a new global metric that effectively quantifies the relative change in policy exploration throughout updates. Building on this metric, we introduce an \textbf{Entropy Ratio Clipping} (ERC) mechanism that imposes bidirectional constraints on the entropy ratio. This stabilizes policy updates at the global distribution level and compensates for the inability of PPO-clip to regulate probability shifts of un-sampled actions. We integrate ERC into both DAPO and GPPO reinforcement learning algorithms. Experiments across multiple benchmarks show that ERC consistently improves performance.