Value-Guided Search for Efficient Chain-of-Thought Reasoning

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wang, Kaiwen, Zhou, Jin Peng, Chang, Jonathan, Gao, Zhaolin, Kallus, Nathan, Brantley, Kianté, Sun, Wen
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866908567742709760
author Wang, Kaiwen
Zhou, Jin Peng
Chang, Jonathan
Gao, Zhaolin
Kallus, Nathan
Brantley, Kianté
Sun, Wen
author_facet Wang, Kaiwen
Zhou, Jin Peng
Chang, Jonathan
Gao, Zhaolin
Kallus, Nathan
Brantley, Kianté
Sun, Wen
contents In this paper, we propose a simple and efficient method for value model training on long-context reasoning traces. Compared to existing process reward models (PRMs), our method does not require a fine-grained notion of "step," which is difficult to define for long-context reasoning models. By collecting a dataset of 2.5 million reasoning traces, we train a 1.5B token-level value model and apply it to DeepSeek models for improved performance with test-time compute scaling. We find that block-wise value-guided search (VGS) with a final weighted majority vote achieves better test-time scaling than standard methods such as majority voting or best-of-n. Moreover, VGS significantly reduces the inference FLOPs required to achieve the same performance of majority voting. Our dataset, model and codebase are open-sourced.
format Preprint
id arxiv_https___arxiv_org_abs_2505_17373
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Value-Guided Search for Efficient Chain-of-Thought Reasoning
Wang, Kaiwen
Zhou, Jin Peng
Chang, Jonathan
Gao, Zhaolin
Kallus, Nathan
Brantley, Kianté
Sun, Wen
Machine Learning
Artificial Intelligence
Computation and Language
In this paper, we propose a simple and efficient method for value model training on long-context reasoning traces. Compared to existing process reward models (PRMs), our method does not require a fine-grained notion of "step," which is difficult to define for long-context reasoning models. By collecting a dataset of 2.5 million reasoning traces, we train a 1.5B token-level value model and apply it to DeepSeek models for improved performance with test-time compute scaling. We find that block-wise value-guided search (VGS) with a final weighted majority vote achieves better test-time scaling than standard methods such as majority voting or best-of-n. Moreover, VGS significantly reduces the inference FLOPs required to achieve the same performance of majority voting. Our dataset, model and codebase are open-sourced.
title Value-Guided Search for Efficient Chain-of-Thought Reasoning
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2505.17373