[论文] SoftCache: Adaptive Step Size Caching for Diffusion Transformers

论文概要

研究领域: CV 作者: Shimon Malnick, Nadav Orenstein, Ohad Fried, Shai Avidan 发布时间: 2026-10-15 arXiv: 2610.22369

中文摘要

扩散和流模型通过迭代步骤生成样本，每一步都是对大型网络的完整前向传播，而这主导了生成成本。免训练缓存通过重用最近评估的激活来加速生成。动态缓存方法通过将网络输入的廉价探针累积到陈旧信号中来决定何时刷新。它们执行硬缓存：信号是刷新和重用之间的二元开关。我们观察到该信号低于阈值的距离（余量）能预测重用引入的误差。我们提出 SoftCache，执行软缓存：让余量决定每个缓存步骤走多远。余量大时，采样器跳过多个预定节点；随着刷新临近，步长逐渐缩小回原始步长。这些自适应跳跃可能落在预定节点之间，使采样器偏离固定网格。SoftCache 无需训练，也不增加每骨干网络的设置：在 FLUX 图像生成和 HunyuanVideo、Wan2.1 视频生成上，在 3.5 倍到 6.9 倍加速比下，在所有四项保真度指标上均优于所有基线。

原文摘要

Diffusion and flow models generate a sample over iterative steps, each a full pass through a large network, and that pass dominates generation cost. Training-free caching accelerates them by reusing activations from recent evaluations instead of recomputing them. Dynamic caching methods decide when to refresh by accumulating a cheap probe of the network input into a staleness signal and spending a fresh evaluation once it crosses a threshold. They perform hard caching: the signal is a binary switch between refresh and reuse, and every step, fresh or cached, advances by the prescribed step size. We observe that the signal carries much more than this switch. How far it sits below the threshold, its headroom, predicts the error induced by reuse. We propose SoftCache, which performs soft cachi...

#论文 #arXiv #CV #小凯