arXiv 2511.096112025-11-18cs.CV

MMaDA-Parallel：面向思维感知编辑与生成的多模态大扩散语言模型

MMaDA-Parallel: Multimodal Large Diffusion Language Models for Thinking-Aware Editing and Generation

Peking University（北京大学）；ByteDance（字节跳动）；Princeton University（普林斯顿大学）；CASIA（中国科学院自动化研究所）；The University of Chicago（芝加哥大学）

Ye Tian, Ling Yang, Jiongfan Yang, Anran Wang, Yu Tian, Jiani Zheng, Haochen Wang, Zhiyang Teng, Zhuochen Wang, Yinjie Wang, Yunhai Tong, Mengdi Wang, Xiangtai Li

备注：: Project Page: https://tyfeld.github.io/mmadaparellel.github.io/

阅读 PDF

arXiv 页面 ↗HTML 原文 ↗

AI总结：

针对现有自回归思维感知生成方法因误差传播导致性能下降的问题，提出并行多模态扩散框架MMaDA-Parallel及配套基准ParaBench，经监督微调与ParaRL优化后，跨模态对齐性能较SOTA模型Bagel提升6.9%。

摘要：

尽管思维感知生成旨在提升复杂任务的性能，我们发现现有串行自回归方法存在一种关键失效模式：由于误差传播，这类方法反而会反常地降低性能。为系统分析该问题，我们提出ParaBench，这一全新基准可同时评估文本与图像两种输出模态。基于ParaBench的分析显示，这种性能下降与生成的推理过程和最终图像间的对齐度不足高度相关。为解决该问题，我们提出并行多模态扩散框架MMaDA-Parallel，可在整个去噪轨迹中实现文本与图像的持续双向交互。MMaDA-Parallel先通过监督微调训练，再通过并行强化学习（ParaRL）进一步优化，这一新颖策略沿轨迹施加语义奖励以保障跨模态一致性。实验验证，我们的模型显著提升了跨模态对齐与语义一致性，在ParaBench上的输出对齐指标较当前最优模型Bagel提升6.9%，为思维感知图像合成建立了更鲁棒的范式。我们的代码已开源在https://github.com/tyfeld/MMaDA-Parallel。

英文摘要：

While thinking-aware generation aims to improve performance on complex tasks, we identify a critical failure mode where existing sequential, autoregressive approaches can paradoxically degrade performance due to error propagation. To systematically analyze this issue, we propose ParaBench, a new benchmark designed to evaluate both text and image output modalities. Our analysis using ParaBench reveals that this performance degradation is strongly correlated with poor alignment between the generated reasoning and the final image. To resolve this, we propose a parallel multimodal diffusion framework, MMaDA-Parallel, that enables continuous, bidirectional interaction between text and images throughout the entire denoising trajectory. MMaDA-Parallel is trained with supervised finetuning and then further optimized by Parallel Reinforcement Learning (ParaRL), a novel strategy that applies semantic rewards along the trajectory to enforce cross-modal consistency. Experiments validate that our model significantly improves cross-modal alignment and semantic consistency, achieving a 6.9\% improvement in Output Alignment on ParaBench compared to the state-of-the-art model, Bagel, establishing a more robust paradigm for thinking-aware image synthesis. Our code is open-sourced at https://github.com/tyfeld/MMaDA-Parallel

机构、总结、中文摘要均由AI分析而来，可能存在错误，请以论文正文为准。