Beyond the Stability-Exploration Dilemma: Environmental Regularization for LLM Policy Optimization

Hugging Face Daily Papers Papers

Summary

The paper introduces ERPO, a method that moves regularization from the action-side to the input-side by controlling query distribution, addressing the stability-exploration dilemma in LLM policy optimization, and showing improvements on mathematical reasoning benchmarks.

Policy optimization (PO) for Large Language Models faces a stability--exploration trade-off, currently mediated by an action-side Policy-KL regularizer. This puts practitioners in a double bind: keeping Policy-KL constrains response behavior and consumes the action-side exploration budget, while dropping it leaves the optimization without an explicit drift control. We argue for an alternative that breaks the dilemma by moving regularization to the input side. As training progresses, the distribution over training queries induced by the current policy drifts unchecked from its pre-RL reference distribution. Concretely, Environment-Regularized Policy Optimization (ERPO) introduces a Query-KL (QKL) term that bounds this query distribution shift, together with a dataset-static reference-derived per-query weight that biases each per-query update toward queries typical under the reference. The QKL gradient flows strictly through the query likelihood; the response score function used by policy-gradient estimators does not appear in the QKL term, so QKL exerts no direct gradient pressure on the response distribution---exploration is preserved. ERPO plugs into GRPO/PPO/REINFORCE-style pipelines without additional forward passes. On six mathematical reasoning benchmarks, ERPO replaces the standard Policy-KL regularizer while achieving effective control over query distribution drift, delivering stronger accuracy and substantially more stable behavior under high-temperature decoding and long-horizon training.Our source code are available at https://github.com/alibaba/ERPO
Original Article
View Cached Full Text

Cached at: 08/25/26, 08:34 AM

Paper page - Beyond the Stability-Exploration Dilemma: Environmental Regularization for LLM Policy Optimization

Source: https://huggingface.co/papers/2608.23311 Published on Aug 24

·

Submitted byhttps://huggingface.co/xinyu1607

heon Aug 25

Abstract

ERPO replaces action-side policy regularization with input-side query distribution control to stabilize reinforcement learning for language models while preserving response exploration.

Policy optimization(PO) for Large Language Models faces a stability--exploration trade-off, currently mediated by an action-sidePolicy-KL regularizer. This puts practitioners in a double bind: keeping Policy-KL constrains response behavior and consumes the action-side exploration budget, while dropping it leaves the optimization without an explicit drift control. We argue for an alternative that breaks the dilemma by moving regularization to the input side. As training progresses, the distribution over training queries induced by the current policy drifts unchecked from its pre-RL reference distribution. Concretely,Environment-Regularized Policy Optimization(ERPO) introduces aQuery-KL(QKL) term that bounds this query distribution shift, together with a dataset-static reference-derived per-query weight that biases each per-query update toward queries typical under the reference. The QKL gradient flows strictly through the query likelihood; the response score function used bypolicy-gradient estimatorsdoes not appear in the QKL term, so QKL exerts no direct gradient pressure on the response distribution---exploration is preserved. ERPO plugs intoGRPO/PPO/REINFORCE-style pipelines without additional forward passes. On six mathematical reasoning benchmarks, ERPO replaces the standardPolicy-KL regularizerwhile achieving effective control over query distribution drift, delivering stronger accuracy and substantially more stable behavior under high-temperature decoding and long-horizon training.Our source code are available at https://github.com/alibaba/ERPO

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2608\.23311

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2608.23311 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2608.23311 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2608.23311 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Revisiting Entropy Regularization: Adaptive Coefficient Unlocks Its Potential for LLM Reinforcement Learning

arXiv cs.CL

This paper proposes Adaptive Entropy Regularization (AER), a framework that dynamically balances exploration and exploitation in LLM reinforcement learning by addressing policy entropy collapse through difficulty-aware coefficient allocation and initial-anchored target entropy. Experiments on mathematical reasoning benchmarks demonstrate consistent improvements in both accuracy and exploration capability.

VERPO: Verified Evidence Regularized Policy Optimization

arXiv cs.LG

VERPO introduces a framework for verified evidence regularized policy optimization in language models, improving performance on scientific reasoning and tool-use tasks by treating evidence as a proposal for policy correction while maintaining outcome objectives.

Rethinking the Divergence Regularization in LLM RL

Hugging Face Daily Papers

This paper introduces DRPO, which replaces the hard mask in DPPO with a smooth advantage-weighted quadratic regularizer to improve stability and efficiency in LLM reinforcement learning by providing continuous gradient corrections beyond trust-region boundaries.

ESPO: Early-Stopping Proximal Policy Optimization

Hugging Face Daily Papers

ESPO introduces an early-stopping mechanism for reinforcement learning that detects and terminates failed reasoning trajectories in LLMs, improving mathematical reasoning performance while reducing compute by over 20%.