Agentic Entropy-Balanced Policy Optimization (AEPO)

This repository hosts the Qwen3-14B-AEPO-DeepSearch model, a checkpoint for Agentic Entropy-Balanced Policy Optimization (AEPO), an agentic Reinforcement Learning (RL) algorithm presented in the paper Agentic Entropy-Balanced Policy Optimization.

Paper Abstract

Recently, Agentic Reinforcement Learning (Agentic RL) has made significant progress in incentivizing the multi-turn, long-horizon tool-use capabilities of web agents. While mainstream agentic RL algorithms autonomously explore high-uncertainty tool-call steps under the guidance of entropy, excessive reliance on entropy signals can impose further constraints, leading to the training collapse. In this paper, we delve into the challenges caused by entropy and propose the Agentic Entropy-Balanced Policy Optimization (AEPO), an agentic RL algorithm designed to balance entropy in both the rollout and policy update phases. AEPO comprises two core components: (1) a dynamic entropy-balanced rollout mechanism that adaptively allocate global and branch sampling budget through entropy pre-monitoring, while imposing a branch penalty on consecutive high-entropy tool-call steps to prevent over-branching issues; and (2) Entropy-Balanced Policy Optimization that inserts a stop-gradient operation into the high-entropy clipping term to preserve and properly rescale gradients on high-entropy tokens, while incorporating entropy-aware advantage estimation to prioritize learning on high-uncertainty tokens. Results across 14 challenging datasets show that AEPO consistently outperforms 7 mainstream RL algorithms. With just 1K RL samples, Qwen3-14B with AEPO achieves impressive results: 47.6% on GAIA, 11.2% on Humanity's Last Exam, and 43.0% on WebWalker for Pass@1; 65.0% on GAIA, 26.0% on Humanity's Last Exam, and 70.0% on WebWalker for Pass@5. Further analysis reveals that AEPO improves rollout sampling diversity while maintaining stable policy entropy, facilitating scalable web agent training.

Code and Resources

The official GitHub repository for this project, including the codebase, datasets, and detailed training and evaluation instructions, can be found here: https://github.com/RUC-NLPIR/ARPO.

Usage

This model is compatible with the Hugging Face transformers library. For detailed instructions on loading and utilizing the model, including its agentic setup and tool integration, please refer to the comprehensive documentation and scripts provided in the official GitHub repository.