Ensembling Large Language Models with Process Reward-Guided Tree Search for Better Complex Reasoning (2412.15797v1)

Abstract: Despite recent advances in LLMs, open-source models often struggle to consistently perform well on complex reasoning tasks. Existing ensemble methods, whether applied at the token or output levels, fail to address these challenges. In response, we present LLM Ensemble with Monte Carlo Tree Search (LE-MCTS), a novel framework for process-level ensembling of LLMs. LE-MCTS formulates step-by-step reasoning with an ensemble of LLMs as a Markov decision process. In this framework, states represent intermediate reasoning paths, while actions consist of generating the next reasoning step using one of the LLMs selected from a predefined pool. Guided by a process-based reward model, LE-MCTS performs a tree search over the reasoning steps generated by different LLMs, identifying the most accurate reasoning chain. Experimental results on five mathematical reasoning benchmarks demonstrate that our approach outperforms both single LLM decoding algorithms and LLM ensemble methods. Notably, LE-MCTS improves performance by 3.6% and 4.3% on the MATH and MQA datasets, respectively, highlighting its effectiveness in solving complex reasoning problems.