Enhancing LLM Reasoning with Reward-guided Tree Search

TL;DR

This paper introduces STILL-1, a reward-guided tree search framework that improves large language models' reasoning abilities, especially in mathematical tasks, by integrating policy, reward models, and search algorithms.

Contribution

The paper presents a novel reward-guided tree search method for enhancing LLM reasoning, with detailed design considerations and extensive evaluation on challenging datasets.

Findings

Significant improvement in mathematical reasoning accuracy.

Effective integration of policy and reward models in tree search.

Enhanced reasoning capabilities demonstrated on four datasets.

Abstract

Recently, test-time scaling has garnered significant attention from the research community, largely due to the substantial advancements of the o1 model released by OpenAI. By allocating more computational resources during the inference phase, large language models~(LLMs) can extensively explore the solution space by generating more thought tokens or diverse solutions, thereby producing more accurate responses. However, developing an o1-like reasoning approach is challenging, and researchers have been making various attempts to advance this open area of research. In this paper, we present a preliminary exploration into enhancing the reasoning abilities of LLMs through reward-guided tree search algorithms. This framework is implemented by integrating the policy model, reward model, and search algorithm. It is primarily constructed around a tree search algorithm, where the policy model navigates a dynamically expanding tree guided by a specially trained reward model. The implemented framework is denoted as \textbf{STILL-1}. We thoroughly explore various design considerations necessary for implementing this framework and provide a detailed report of the technical aspects. To assess the effectiveness of our approach, we focus on mathematical reasoning tasks and conduct extensive evaluations on four challenging datasets, significantly enhancing the reasoning abilities of LLMs.