Co-GRPO: Co-Optimized Group Relative Policy Optimization for Masked Diffusion Model

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Zhou, Renping, Ni, Zanlin, Chen, Tianyi, Liu, Zeyu, Yue, Yang, Wang, Yulin, Wang, Yuxuan, Liu, Jingshu, Huang, Gao
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866909976681775104
author Zhou, Renping
Ni, Zanlin
Chen, Tianyi
Liu, Zeyu
Yue, Yang
Wang, Yulin
Wang, Yuxuan
Liu, Jingshu
Huang, Gao
author_facet Zhou, Renping
Ni, Zanlin
Chen, Tianyi
Liu, Zeyu
Yue, Yang
Wang, Yulin
Wang, Yuxuan
Liu, Jingshu
Huang, Gao
contents Recently, Masked Diffusion Models (MDMs) have shown promising potential across vision, language, and cross-modal generation. However, a notable discrepancy exists between their training and inference procedures. In particular, MDM inference is a multi-step, iterative process governed not only by the model itself but also by various schedules that dictate the token-decoding trajectory (e.g., how many tokens to decode at each step). In contrast, MDMs are typically trained using a simplified, single-step BERT-style objective that masks a subset of tokens and predicts all of them simultaneously. This step-level simplification fundamentally disconnects the training paradigm from the trajectory-level nature of inference, leaving the inference schedules never optimized during training. In this paper, we introduce Co-GRPO, which reformulates MDM generation as a unified Markov Decision Process (MDP) that jointly incorporates both the model and the inference schedule. By applying Group Relative Policy Optimization at the trajectory level, Co-GRPO cooperatively optimizes model parameters and schedule parameters under a shared reward, without requiring costly backpropagation through the multi-step generation process. This holistic optimization aligns training with inference more thoroughly and substantially improves generation quality. Empirical results across four benchmarks-ImageReward, HPS, GenEval, and DPG-Bench-demonstrate the effectiveness of our approach. For more details, please refer to our project page: https://co-grpo.github.io/ .
format Preprint
id arxiv_https___arxiv_org_abs_2512_22288
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Co-GRPO: Co-Optimized Group Relative Policy Optimization for Masked Diffusion Model
Zhou, Renping
Ni, Zanlin
Chen, Tianyi
Liu, Zeyu
Yue, Yang
Wang, Yulin
Wang, Yuxuan
Liu, Jingshu
Huang, Gao
Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
Recently, Masked Diffusion Models (MDMs) have shown promising potential across vision, language, and cross-modal generation. However, a notable discrepancy exists between their training and inference procedures. In particular, MDM inference is a multi-step, iterative process governed not only by the model itself but also by various schedules that dictate the token-decoding trajectory (e.g., how many tokens to decode at each step). In contrast, MDMs are typically trained using a simplified, single-step BERT-style objective that masks a subset of tokens and predicts all of them simultaneously. This step-level simplification fundamentally disconnects the training paradigm from the trajectory-level nature of inference, leaving the inference schedules never optimized during training. In this paper, we introduce Co-GRPO, which reformulates MDM generation as a unified Markov Decision Process (MDP) that jointly incorporates both the model and the inference schedule. By applying Group Relative Policy Optimization at the trajectory level, Co-GRPO cooperatively optimizes model parameters and schedule parameters under a shared reward, without requiring costly backpropagation through the multi-step generation process. This holistic optimization aligns training with inference more thoroughly and substantially improves generation quality. Empirical results across four benchmarks-ImageReward, HPS, GenEval, and DPG-Bench-demonstrate the effectiveness of our approach. For more details, please refer to our project page: https://co-grpo.github.io/ .
title Co-GRPO: Co-Optimized Group Relative Policy Optimization for Masked Diffusion Model
topic Machine Learning
Artificial Intelligence
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2512.22288