TC-Pade：面向扩散加速的轨迹一致 Padé 逼近
TC-Pade: Trajectory-Consistent Pade Approximation for Diffusion Acceleration
用有理函数（Padé 逼近）替代 Taylor 类多项式外推来做扩散模型的特征缓存预测，并配合自适应系数调制与分阶段预测策略，可直接嵌入 DiT/FLUX/Wan 等视觉生成骨干，在 20-30 步低步数区间实现约 2-3 倍采样加速而不掉画质。
原文依据 we propose Trajectory-Consistent Pade(TC-Pade) approximation, a feature prediction framework grounded in Pade approximation
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论尽管扩散模型达到了最优的生成质量，但其迭代采样过程带来的巨大计算负担仍制约着其应用。特征缓存技术在高步数（如 50 步）下能实现有效加速，但在 20-30 步这一实际常用的低步数区间却表现出严重局限。随着步间间隔增大，TaylorSeer 这类基于多项式的外推器会遭受误差累积与轨迹漂移。同时，传统缓存策略往往忽略了不同去噪阶段各自不同的动力学特性。为应对这些挑战，我们提出轨迹一致 Padé（TC-Pade）逼近，一个基于 Padé 逼近的特征预测框架。通过用有理函数对特征演化建模，我们的方法比基于泰勒的方法更准确地刻画渐近行为与过渡行为。为在降低步数的情况下实现稳定且轨迹一致的采样，TC-Pade 引入了：(1) 自适应系数调制，利用历史缓存残差来检测细微的轨迹转变；(2) 步数感知的预测策略，针对采样早期、中期与晚期各自不同的动力学特性量身定制。在 DiT-XL/2、FLUX.1-dev 与 Wan2.1 上跨图像与视频生成的广泛实验证明了 TC-Pade 的有效性。例如，TC-Pade 在 FLUX.1-dev 上实现 2.88 倍加速、在 Wan2.1 上实现 1.72 倍加速，同时在 FID、CLIP、Aesthetic 与 VBench-2.0 指标上保持高质量，显著优于现有的特征缓存方法。
英文原文摘要 1630 字符
Despite achieving state-of-the-art generation quality, diffusion models are hindered by the substantial computational burden of their iterative sampling process. While feature caching techniques achieve effective acceleration at higher step counts (e.g., 50 steps), they exhibit critical limitations in the practical low-step regime of 20-30 steps. As the interval between steps increases, polynomial-based extrapolators like TaylorSeer suffer from error accumulation and trajectory drift. Meanwhile, conventional caching strategies often overlook the distinct dynamical properties of different denoising phases. To address these challenges, we propose Trajectory-Consistent Pade(TC-Pade) approximation, a feature prediction framework grounded in Pade approximation. By modeling feature evolution through rational functions, our approach captures asymptotic and transitional behaviors more accurately than Taylor-based methods. To enable stable and trajectory-consistent sampling under reduced step counts, TC-Pade incorporates (1) adaptive coefficient modulation that leverages historical cached residuals to detect subtle trajectory transitions, and (2) step-aware prediction strategies tailored to the distinct dynamics of early, mid, and late sampling stages. Extensive experiments on DiT-XL/2, FLUX.1-dev, and Wan2.1 across both image and video generation demonstrate the effectiveness of TC-Pade. For instance, TC-Pade achieves 2.88xacceleration on FLUX.1-dev and 1.72xon Wan2.1 while maintaining high quality across FID, CLIP, Aesthetic, and VBench-2.0 metrics, substantially outperforming existing feature caching methods.