URPO: A Unified Reward & Policy Optimization Framework for Large Language Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Lu, Songshuo, Wang, Hua, Chen, Zhi, Tang, Yaohua
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913956454465536
author Lu, Songshuo
Wang, Hua
Chen, Zhi
Tang, Yaohua
author_facet Lu, Songshuo
Wang, Hua
Chen, Zhi
Tang, Yaohua
contents Large-scale alignment pipelines typically pair a policy model with a separately trained reward model whose parameters remain frozen during reinforcement learning (RL). This separation creates a complex, resource-intensive pipeline and suffers from a performance ceiling due to a static reward signal. We propose a novel framework, Unified Reward & Policy Optimization (URPO), that unifies instruction-following ("player") and reward modeling ("referee") within a single model and a single training phase. Our method recasts all alignment data-including preference pairs, verifiable reasoning, and open-ended instructions-into a unified generative format optimized by a single Group-Relative Policy Optimization (GRPO) loop. This enables the model to learn from ground-truth preferences and verifiable logic while simultaneously generating its own rewards for open-ended tasks. Experiments on the Qwen2.5-7B model demonstrate URPO's superiority. Our unified model significantly outperforms a strong baseline using a separate generative reward model, boosting the instruction-following score on AlpacaEval from 42.24 to 44.84 and the composite reasoning average from 32.66 to 35.66. Furthermore, URPO cultivates a superior internal evaluator as a byproduct of training, achieving a RewardBench score of 85.15 and surpassing the dedicated reward model it replaces (83.55). By eliminating the need for a separate reward model and fostering a co-evolutionary dynamic between generation and evaluation, URPO presents a simpler, more efficient, and more effective path towards robustly aligned language models.
format Preprint
id arxiv_https___arxiv_org_abs_2507_17515
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle URPO: A Unified Reward & Policy Optimization Framework for Large Language Models
Lu, Songshuo
Wang, Hua
Chen, Zhi
Tang, Yaohua
Computer Vision and Pattern Recognition
Computation and Language
Large-scale alignment pipelines typically pair a policy model with a separately trained reward model whose parameters remain frozen during reinforcement learning (RL). This separation creates a complex, resource-intensive pipeline and suffers from a performance ceiling due to a static reward signal. We propose a novel framework, Unified Reward & Policy Optimization (URPO), that unifies instruction-following ("player") and reward modeling ("referee") within a single model and a single training phase. Our method recasts all alignment data-including preference pairs, verifiable reasoning, and open-ended instructions-into a unified generative format optimized by a single Group-Relative Policy Optimization (GRPO) loop. This enables the model to learn from ground-truth preferences and verifiable logic while simultaneously generating its own rewards for open-ended tasks. Experiments on the Qwen2.5-7B model demonstrate URPO's superiority. Our unified model significantly outperforms a strong baseline using a separate generative reward model, boosting the instruction-following score on AlpacaEval from 42.24 to 44.84 and the composite reasoning average from 32.66 to 35.66. Furthermore, URPO cultivates a superior internal evaluator as a byproduct of training, achieving a RewardBench score of 85.15 and surpassing the dedicated reward model it replaces (83.55). By eliminating the need for a separate reward model and fostering a co-evolutionary dynamic between generation and evaluation, URPO presents a simpler, more efficient, and more effective path towards robustly aligned language models.
title URPO: A Unified Reward & Policy Optimization Framework for Large Language Models
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2507.17515