Process Reward Models That Think

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Khalifa, Muhammad, Agarwal, Rishabh, Logeswaran, Lajanugen, Kim, Jaekyeom, Peng, Hao, Lee, Moontae, Lee, Honglak, Wang, Lu
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917129915203584
author Khalifa, Muhammad
Agarwal, Rishabh
Logeswaran, Lajanugen
Kim, Jaekyeom
Peng, Hao
Lee, Moontae
Lee, Honglak
Wang, Lu
author_facet Khalifa, Muhammad
Agarwal, Rishabh
Logeswaran, Lajanugen
Kim, Jaekyeom
Peng, Hao
Lee, Moontae
Lee, Honglak
Wang, Lu
contents Step-by-step verifiers -- also known as process reward models (PRMs) -- are a key ingredient for test-time scaling. PRMs require step-level supervision, making them expensive to train. This work aims to build data-efficient PRMs as verbalized step-wise reward models that verify every step in the solution by generating a verification chain-of-thought (CoT). We propose ThinkPRM, a long CoT verifier fine-tuned on orders of magnitude fewer process labels than those required by discriminative PRMs. Our approach capitalizes on the inherent reasoning abilities of long CoT models, and outperforms LLM-as-a-Judge and discriminative verifiers -- using only 1% of the process labels in PRM800K -- across several challenging benchmarks. Specifically, ThinkPRM beats the baselines on ProcessBench, MATH-500, and AIME '24 under best-of-N selection and reward-guided search. In an out-of-domain evaluation on a subset of GPQA-Diamond and LiveCodeBench, our PRM surpasses discriminative verifiers trained on the full PRM800K by 8% and 4.5%, respectively. Lastly, under the same token budget, ThinkPRM scales up verification compute more effectively compared to LLM-as-a-Judge, outperforming it by 7.2% on a subset of ProcessBench. Our work highlights the value of generative, long CoT PRMs that can scale test-time compute for verification while requiring minimal supervision for training. Our code, data, and models are released at https://github.com/mukhal/thinkprm.
format Preprint
id arxiv_https___arxiv_org_abs_2504_16828
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Process Reward Models That Think
Khalifa, Muhammad
Agarwal, Rishabh
Logeswaran, Lajanugen
Kim, Jaekyeom
Peng, Hao
Lee, Moontae
Lee, Honglak
Wang, Lu
Machine Learning
Artificial Intelligence
Computation and Language
Step-by-step verifiers -- also known as process reward models (PRMs) -- are a key ingredient for test-time scaling. PRMs require step-level supervision, making them expensive to train. This work aims to build data-efficient PRMs as verbalized step-wise reward models that verify every step in the solution by generating a verification chain-of-thought (CoT). We propose ThinkPRM, a long CoT verifier fine-tuned on orders of magnitude fewer process labels than those required by discriminative PRMs. Our approach capitalizes on the inherent reasoning abilities of long CoT models, and outperforms LLM-as-a-Judge and discriminative verifiers -- using only 1% of the process labels in PRM800K -- across several challenging benchmarks. Specifically, ThinkPRM beats the baselines on ProcessBench, MATH-500, and AIME '24 under best-of-N selection and reward-guided search. In an out-of-domain evaluation on a subset of GPQA-Diamond and LiveCodeBench, our PRM surpasses discriminative verifiers trained on the full PRM800K by 8% and 4.5%, respectively. Lastly, under the same token budget, ThinkPRM scales up verification compute more effectively compared to LLM-as-a-Judge, outperforming it by 7.2% on a subset of ProcessBench. Our work highlights the value of generative, long CoT PRMs that can scale test-time compute for verification while requiring minimal supervision for training. Our code, data, and models are released at https://github.com/mukhal/thinkprm.
title Process Reward Models That Think
topic Machine Learning
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2504.16828