Unsupervised Process Reward Models

Hugging Face Daily Papers Papers

Summary

This paper proposes unsupervised Process Reward Models (uPRM) that eliminate the need for human annotations by using LLM next-token probabilities to identify erroneous reasoning steps, achieving up to 15% accuracy improvements over LLM-as-a-Judge and performing comparably to supervised PRMs as verifiers and reward signals.

Process Reward Models (PRMs) are a powerful mechanism for steering large language model reasoning by providing fine-grained, step-level supervision. However, this effectiveness comes at a significant cost: PRMs require expert annotations for every reasoning step, making them costly and difficult to scale. Here, we propose a method for training unsupervised PRMs (uPRM) that requires no human supervision, neither at the level of step-by-step annotations nor through ground-truth verification of final answers. The key idea behind our approach is to define a scoring function, derived from LLM next-token probabilities, that jointly assesses candidate positions of first erroneous steps across a batch of reasoning trajectories. We demonstrate the effectiveness of uPRM across diverse scenarios: (i) uPRM achieves up to 15% absolute accuracy improvements over the LLM-as-a-Judge in identifying first erroneous steps on the ProcessBench dataset; (ii) as a verifier for test-time scaling, uPRM performs comparably to supervised PRMs and outperforms the majority voting baseline by up to 6.9%, and (iii) when used as a reward signal in reinforcement learning, uPRM enables more robust policy optimization throughout training compared to a supervised PRM trained using ground-truth labels. Overall, our results open a path toward scalable reward modeling for complex reasoning tasks.
Original Article
View Cached Full Text

Cached at: 05/22/26, 10:19 AM

Paper page - Unsupervised Process Reward Models

Source: https://huggingface.co/papers/2605.10158

Abstract

Unsupervised reward models eliminate the need for human annotations in training by leveraging language model next-token probabilities to identify erroneous reasoning steps and improve policy optimization in reinforcement learning.

Process Reward Models(PRMs) are a powerful mechanism for steering largelanguage modelreasoning by providing fine-grained, step-level supervision. However, this effectiveness comes at a significant cost: PRMs require expert annotations for every reasoning step, making them costly and difficult to scale. Here, we propose a method for training unsupervised PRMs (uPRM) that requires no human supervision, neither at the level of step-by-step annotations nor through ground-truth verification of final answers. The key idea behind our approach is to define a scoring function, derived from LLMnext-token probabilities, that jointly assesses candidate positions offirst erroneous stepsacross a batch ofreasoning trajectories. We demonstrate the effectiveness of uPRM across diverse scenarios: (i) uPRM achieves up to 15% absolute accuracy improvements over the LLM-as-a-Judge in identifyingfirst erroneous stepson theProcessBenchdataset; (ii) as a verifier fortest-time scaling, uPRM performs comparably to supervised PRMs and outperforms the majority voting baseline by up to 6.9%, and (iii) when used as a reward signal inreinforcement learning, uPRM enables more robustpolicy optimizationthroughout training compared to a supervised PRM trained using ground-truth labels. Overall, our results open a path toward scalable reward modeling for complex reasoning tasks.

View arXiv pageView PDFAdd to collection

Get this paper in your agent:

hf papers read 2605\.10158

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.10158 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.10158 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.10158 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Process Rewards with Learned Reliability

arXiv cs.CL

BetaPRM is a process reward model that predicts both a step-level success probability and the reliability of that prediction using a Beta belief from Monte Carlo continuations, enabling adaptive computation allocation that reduces token usage by up to 33.57% while improving accuracy.

SCI-PRM: A Tool Aware Process Reward Model for Scientific Reasoning Verification

arXiv cs.AI

SCI-PRM introduces a tool-aware Process Reward Model for scientific reasoning, trained on the SCIPRM70K dataset featuring 'Chain-of-Tool' trajectories that interleave reasoning with scientific tool execution. It enables effective test-time scaling and serves as a dense reward signal in reinforcement learning, outperforming proprietary models like GPT-5-Mini on tool-calling steps across scientific benchmarks.