TreeRL: LLM Reinforcement Learning with On-Policy Tree Search
This addresses the challenge of efficient and effective RL training for LLMs in reasoning tasks, offering a novel approach that could benefit AI researchers and practitioners, though it appears incremental by building on existing tree search techniques.
The paper tackles the problem of improving reinforcement learning for large language models by proposing TreeRL, a framework that incorporates on-policy tree search to enhance exploration and provide dense process rewards, eliminating the need for a separate reward model. It demonstrates superior performance on math and code reasoning benchmarks compared to traditional methods.
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.