Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Yao, Jiashu, Huang, Heyan, Luo, Chuwei, Wu, Daiqing, Liu, Zeming, Guo, Yuhang, Kang, Yangyang
Format: Preprint
Veröffentlicht: 2026
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2604.11510
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915934761910272
author Yao, Jiashu
Huang, Heyan
Luo, Chuwei
Wu, Daiqing
Liu, Zeming
Guo, Yuhang
Kang, Yangyang
author_facet Yao, Jiashu
Huang, Heyan
Luo, Chuwei
Wu, Daiqing
Liu, Zeming
Guo, Yuhang
Kang, Yangyang
contents To encourage diverse exploration in reinforcement learning (RL) for large language models (LLMs) without compromising accuracy, we propose Policy Split, a novel paradigm that bifurcates the policy into normal and high-entropy modes with a high-entropy prompt. While sharing model parameters, the two modes undergo collaborative dual-mode entropy regularization tailored to distinct objectives. Specifically, the normal mode optimizes for task correctness, while the high-entropy mode incorporates a preference for exploration, and the two modes learn collaboratively. Extensive experiments demonstrate that our approach consistently outperforms established entropy-guided RL baselines across various model sizes in general and creative tasks. Further analysis reveals that Policy Split facilitates dual-mode exploration, where the high-entropy mode generates distinct behavioral patterns to the normal mode, providing unique learning signals.
format Preprint
id arxiv_https___arxiv_org_abs_2604_11510
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Policy Split: Incentivizing Dual-Mode Exploration in LLM Reinforcement with Dual-Mode Entropy Regularization
Yao, Jiashu
Huang, Heyan
Luo, Chuwei
Wu, Daiqing
Liu, Zeming
Guo, Yuhang
Kang, Yangyang
Computation and Language
Artificial Intelligence
Machine Learning
To encourage diverse exploration in reinforcement learning (RL) for large language models (LLMs) without compromising accuracy, we propose Policy Split, a novel paradigm that bifurcates the policy into normal and high-entropy modes with a high-entropy prompt. While sharing model parameters, the two modes undergo collaborative dual-mode entropy regularization tailored to distinct objectives. Specifically, the normal mode optimizes for task correctness, while the high-entropy mode incorporates a preference for exploration, and the two modes learn collaboratively. Extensive experiments demonstrate that our approach consistently outperforms established entropy-guided RL baselines across various model sizes in general and creative tasks. Further analysis reveals that Policy Split facilitates dual-mode exploration, where the high-entropy mode generates distinct behavioral patterns to the normal mode, providing unique learning signals.
title Policy Split: Incentivizing Dual-Mode Exploration in LLM Reinforcement with Dual-Mode Entropy Regularization
topic Computation and Language
Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2604.11510