Salvato in:
Dettagli Bibliografici
Autori principali: Hu, Xinyu, He, Yancheng, Wang, Weixun, Feng, Tao, Lin, Li, Liu, Jiashun, Su, Wenbo, Zheng, Bo, Wan, Xiaojun
Natura: Preprint
Pubblicazione: 2026
Soggetti:
Accesso online:https://arxiv.org/abs/2601.20327
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912865966882816
author Hu, Xinyu
He, Yancheng
Wang, Weixun
Feng, Tao
Lin, Li
Liu, Jiashun
Su, Wenbo
Zheng, Bo
Wan, Xiaojun
author_facet Hu, Xinyu
He, Yancheng
Wang, Weixun
Feng, Tao
Lin, Li
Liu, Jiashun
Su, Wenbo
Zheng, Bo
Wan, Xiaojun
contents Automatic evaluation is crucial yet challenging for open-ended natural language generation, especially when rule-based metrics are infeasible. Compared with traditional methods, the recent LLM-as-a-Judge paradigms enable better and more flexible evaluation, and show promise as generative reward models for reinforcement learning. However, prior work has revealed a notable gap between their seemingly impressive benchmark performance and actual effectiveness in RL practice. We attribute this issue to some limitations in existing studies, including the dominance of pairwise evaluation and inadequate optimization of evaluation criteria. Therefore, we propose CE-RM-4B, a pointwise generative reward model trained with a dedicated two-stage rollout method, and adopting unified query-based criteria. Using only about 5.7K high-quality data curated from the open-source preference dataset, our CE-RM-4B achieves superior performance on diverse reward model benchmarks, especially in Best-of-N scenarios, and delivers more effective improvements in downstream RL practice.
format Preprint
id arxiv_https___arxiv_org_abs_2601_20327
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle CE-RM: A Pointwise Generative Reward Model Optimized via Two-Stage Rollout and Unified Criteria
Hu, Xinyu
He, Yancheng
Wang, Weixun
Feng, Tao
Lin, Li
Liu, Jiashun
Su, Wenbo
Zheng, Bo
Wan, Xiaojun
Computation and Language
Automatic evaluation is crucial yet challenging for open-ended natural language generation, especially when rule-based metrics are infeasible. Compared with traditional methods, the recent LLM-as-a-Judge paradigms enable better and more flexible evaluation, and show promise as generative reward models for reinforcement learning. However, prior work has revealed a notable gap between their seemingly impressive benchmark performance and actual effectiveness in RL practice. We attribute this issue to some limitations in existing studies, including the dominance of pairwise evaluation and inadequate optimization of evaluation criteria. Therefore, we propose CE-RM-4B, a pointwise generative reward model trained with a dedicated two-stage rollout method, and adopting unified query-based criteria. Using only about 5.7K high-quality data curated from the open-source preference dataset, our CE-RM-4B achieves superior performance on diverse reward model benchmarks, especially in Best-of-N scenarios, and delivers more effective improvements in downstream RL practice.
title CE-RM: A Pointwise Generative Reward Model Optimized via Two-Stage Rollout and Unified Criteria
topic Computation and Language
url https://arxiv.org/abs/2601.20327