LESA：面向扩散模型加速的可学习阶段感知预测器
LESA: Learnable Stage-Aware Predictors for Diffusion Model Acceleration
把扩散采样中的特征缓存从「简单复用/免训练外推」升级为「按噪声阶段分专家、用 KAN 学习时序特征映射」的可训练预测器，可直接套用到 DiT 类图像/视频生成模型的推理加速上，在 FLUX、Qwen-Image、HunyuanVideo 上验证了 5–6 倍加速且质量几乎不掉。
原文依据 we propose a LEarnable Stage-Aware (LESA) predictor framework based on two-stage training
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论扩散模型在图像与视频生成任务中已取得显著成功。然而，扩散 Transformer（DiT）的高计算需求给其实际部署带来了重大挑战。特征缓存是一种有前景的加速策略，但现有基于简单复用或免训练预测的方法难以适应扩散过程复杂且依赖阶段的动态特性，往往导致质量下降，并且无法与标准去噪过程保持一致。为解决这一问题，我们提出了基于两阶段训练的可学习阶段感知（LESA）预测器框架。该方法利用 Kolmogorov-Arnold 网络（KAN）从数据中准确学习时序特征映射。我们进一步引入多阶段、多专家架构，为不同噪声水平阶段分配专门的预测器，从而实现更精确、更稳健的特征预测。大量实验表明，我们的方法在实现显著加速的同时保持了高保真生成。实验显示，在 FLUX.1-dev 上实现 5.00 倍加速且质量下降极小（仅 1.0%），在 Qwen-Image 上实现 6.25 倍加速，相比此前最优方法（TaylorSeer）质量提升 20.2%，在 HunyuanVideo 上实现 5.00 倍加速，相比 TaylorSeer 峰值信噪比提升 24.7%。在文本到图像与文本到视频合成任务上取得的最优性能，验证了该基于训练的框架在不同模型上的有效性与泛化能力。我们的代码已包含在补充材料中，并将发布于 GitHub。
英文原文摘要 1631 字符
Diffusion models have achieved remarkable success in image and video generation tasks. However, the high computational demands of Diffusion Transformers (DiTs) pose a significant challenge to their practical deployment. While feature caching is a promising acceleration strategy, existing methods based on simple reusing or training-free forecasting struggle to adapt to the complex, stage-dependent dynamics of the diffusion process, often resulting in quality degradation and failing to maintain consistency with the standard denoising process. To address this, we propose a LEarnable Stage-Aware (LESA) predictor framework based on two-stage training. Our approach leverages a Kolmogorov-Arnold Network (KAN) to accurately learn temporal feature mappings from data. We further introduce a multi-stage, multi-expert architecture that assigns specialized predictors to different noise-level stages, enabling more precise and robust feature forecasting. Extensive experiments show our method achieves significant acceleration while maintaining high-fidelity generation. Experiments demonstrate 5.00xacceleration on FLUX.1-dev with minimal quality degradation (1.0% drop), 6.25xspeedup on Qwen-Image with a 20.2% quality improvement over the previous SOTA (TaylorSeer), and 5.00xacceleration on HunyuanVideo with a 24.7% PSNR improvement over TaylorSeer. State-of-the-art performance on both text-to-image and text-to-video synthesis validates the effectiveness and generalization capability of our training-based framework across different models. Our code is included in the supplementary materials and will be released on GitHub.