TreeRL: LLM Reinforcement Learning with On-Policy Tree Search

TL;DR

TreeRL introduces an on-policy tree search reinforcement learning framework for large language models, enhancing reasoning capabilities and efficiency without needing separate reward models, and demonstrates superior performance on math and code benchmarks.

Contribution

It presents a novel on-policy tree search method integrated into RL training for LLMs, improving exploration and efficiency over existing approaches.

Findings

Outperforms ChainRL on math and code benchmarks

Achieves higher search efficiency with strategic branching

Eliminates need for separate reward model training

Abstract

Reinforcement learning (RL) with tree search has demonstrated superior performance in traditional reasoning tasks. Compared to conventional independent chain sampling strategies with outcome supervision, tree search enables better exploration of the reasoning space and provides dense, on-policy process rewards during RL training but remains under-explored in On-Policy LLM RL. We propose TreeRL, a reinforcement learning framework that directly incorporates on-policy tree search for RL training. Our approach includes intermediate supervision and eliminates the need for a separate reward model training. Existing approaches typically train a separate process reward model, which can suffer from distribution mismatch and reward hacking. We also introduce a cost-effective tree search approach that achieves higher search efficiency under the same generation token budget by strategically branching from high-uncertainty intermediate steps rather than using random branching. Experiments on challenging math and code reasoning benchmarks demonstrate that TreeRL achieves superior performance compared to traditional ChainRL, highlighting the potential of tree search for LLM. TreeRL is open-sourced at https://github.com/THUDM/TreeRL.