arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2604.01552cs.LG
收藏

ZEUS:仅使用二阶预测器加速扩散模型

ZEUS: Accelerating Diffusion Models with Only Second-Order Predictor

Department of Electrical and Computer Engineering, Duke University(杜克大学电气与计算机工程系);Department of Computer Science, Duke University(杜克大学计算机科学系);Department of Statistical Science, Duke University(杜克大学统计科学系)

Yixiao Wang, Ting Jiang, Zishan Shao, Hancheng Ye, Jingwei Sun, Mingyuan Ma, Jianyi Zhang, Yiran Chen, Hai Li

AI总结:

本文提出ZEUS方法,通过二阶预测器减少去噪器调用,结合交错方案稳定连续跳过,实现高效加速,提升速度-保真度性能,达到3.2倍端到端提速。

摘要:

去噪生成模型虽能生成高质量内容,但采样过程中的多次迭代去噪器调用导致推理延迟成为瓶颈。训练免费加速方法通过稀疏化模型架构或缩短采样轨迹来减少延迟。当前方法过于复杂:高阶预测器在激进加速下放大误差,架构修改阻碍部署。超过2倍加速后,步跳创造结构稀缺性——每个局部窗口最多一次新鲜评估,使计算输出及其反向差作为唯一因果信息。基于此,我们提出ZEUS,一种通过二阶预测器预测减少去噪器评估的加速方法,并通过交错方案稳定连续跳过,避免连续外推。ZEUS几乎不增加开销,无特征缓存,无架构修改,兼容不同主干、预测目标和求解器选择。在图像和视频生成中,ZEUS在最近的训练免费基线中一致提升了速度-保真度性能,实现高达3.2倍端到端提速,同时保持感知质量。代码可在https://github.com/Ting-Justin-Jiang/ZEUS获取。

英文摘要:

Denoising generative models deliver high-fidelity generation but remain bottlenecked by inference latency due to the many iterative denoiser calls required during sampling. Training-free acceleration methods reduce latency by either sparsifying the model architecture or shortening the sampling trajectory. Current training-free acceleration methods are more complex than necessary: higher-order predictors amplify error under aggressive speedups, and architectural modifications hinder deployment. Beyond 2x acceleration, step skipping creates structural scarcity -- at most one fresh evaluation per local window -- leaving the computed output and its backward difference as the only causally grounded information. Based on this, we propose ZEUS, an acceleration method that predicts reduced denoiser evaluations using a second-order predictor, and stabilizes aggressive consecutive skipping with an interleaved scheme that avoids back-to-back extrapolations. ZEUS adds essentially zero overhead, no feature caches, and no architectural modifications, and it is compatible with different backbones, prediction objectives, and solver choices. Across image and video generation, ZEUS consistently improves the speed-fidelity performance over recent training-free baselines, achieving up to 3.2x end-to-end speedup while maintaining perceptual quality. Our code is available at: https://github.com/Ting-Justin-Jiang/ZEUS.

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。

↑