发表机构
Shanghai Jiao Tong University; Alibaba Cloud Computing; Alibaba Token Hub, Alibaba Group（上海交通大学; 阿里云计算; 阿里巴巴集团阿里Token Hub）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
MORCA通过离线到在线强化学习训练缓存调度框架，在用户指定加速目标下做出潜在感知的复用/重计算决策，显著提升视频扩散生成的保真度。
AI 中文摘要
扩散变换器（DiTs）在视频合成中取得了显著性能，但其迭代去噪过程面临高推理延迟。为解决这一问题，缓存利用去噪过程中的跨步冗余成为一种有效的加速策略。现有的动态缓存方法通常估计每个去噪步骤中缓存复用引入的误差（步误差）来指导缓存决策，而我们的关注点在于缓存复用对最终生成视频造成的质量损失（终端误差）。我们证明步误差与终端误差并不直接对应，且潜在信息有助于捕捉它们之间的关系，从而为缓存决策提供依据。此外，现有的基于阈值的方法无法提供精确的加速控制，难以满足用户指定加速目标的实际需求。为解决这些局限性，我们提出MORCA，一种通过离线到在线强化学习训练的缓存调度框架，在用户指定的加速目标下做出潜在感知的复用/重计算决策。在多个目标加速比下对不同视频生成模型的广泛实验表明，在可比的计算预算下，MORCA比最先进的缓存方法实现了更好的生成保真度。代码可在以下网址获取：此https URL。
英文摘要
Diffusion Transformers (DiTs) achieve remarkable performance in video synthesis, but their iterative denoising process suffers from high inference latency. To address this, caching has emerged as an effective acceleration strategy by capitalizing on inter-step redundancy during denoising. Existing dynamic caching methods typically estimate the error that cache reuse would introduce at each denoising step (step error) to guide cache decisions, whereas our concern is how much quality loss cache reuse would cause in the final generated video (terminal error). We show that step error does not directly correspond to terminal error and that latent information helps capture their relationship, thereby informing cache decisions. Moreover, existing threshold-based methods cannot provide precise speedup control, making it difficult to meet practical requirements for user-specified acceleration targets. To address these limitations, we introduce MORCA, a cache scheduling framework trained through offline-to-online reinforcement learning to make latent-aware reuse/recompute decisions under user-specified acceleration targets. Extensive experiments on different video generation models across multiple target acceleration ratios demonstrate that MORCA achieves better generation fidelity than state-of-the-art caching methods under comparable computational budgets. Code is available at https://github.com/x10ngyx/MORCA.
Comments22 pages, 8 figures
