SmoothCache: A Universal Inference Acceleration Technique for Diffusion Transformers

TL;DR

SmoothCache is a universal inference acceleration method for Diffusion Transformers that adaptively caches features, significantly speeding up generation while preserving or enhancing quality across multiple modalities.

Contribution

It introduces a model-agnostic caching technique that leverages layer output similarities to accelerate inference in Diffusion Transformers across diverse tasks.

Findings

Achieves 8% to 71% speedup in inference

Maintains or improves generation quality

Effective across image, video, and audio modalities

Abstract

Diffusion Transformers (DiT) have emerged as powerful generative models for various tasks, including image, video, and speech synthesis. However, their inference process remains computationally expensive due to the repeated evaluation of resource-intensive attention and feed-forward modules. To address this, we introduce SmoothCache, a model-agnostic inference acceleration technique for DiT architectures. SmoothCache leverages the observed high similarity between layer outputs across adjacent diffusion timesteps. By analyzing layer-wise representation errors from a small calibration set, SmoothCache adaptively caches and reuses key features during inference. Our experiments demonstrate that SmoothCache achieves 8% to 71% speed up while maintaining or even improving generation quality across diverse modalities. We showcase its effectiveness on DiT-XL for image generation, Open-Sora for text-to-video, and Stable Audio Open for text-to-audio, highlighting its potential to enable real-time applications and broaden the accessibility of powerful DiT models.