Local Ai
Memory-Efficient Training-Free Acceleration of Diffusion Transformers with BaryCache
arXiv:2608.28670v1 Announce Type: new Abstract: Diffusion Transformers achieve high-fidelity image and video generation, but their iterative sampling remains expensive, for each denoising step require
arXiv:2608.28670v1 Announce Type: new Abstract: Diffusion Transformers achieve high-fidelity image and video generation, but their iterative sampling remains expensive, for each denoising step requires large matrix operations. Existing cache-based acceleration reduces redundant computation yet increases the VRAM footprint by storing intermediate states, which can directly constrain inference batch size. In this work, we propose a training-free acceleration method that performs stepwise forecasting for DiT sampling using a Barycentric Extrapolator. By leveraging barycentric extrapolation, our predictor is numerically stable and alleviates oscillatory artifacts analogous to the Runge phenomenon during forward forecasting. Across extensive experiments on both image and video generation, our approach provides a favorable trade-off between memory usage and perceptual quality, while delivering up to 3.30x end-to-end sampling speedup compared with baseline DiT inference.
Related
- SnapGen++: Unleashing Diffusion Transformers for Efficient High-Fidelity Image Generation on Edge Devices
- FeatFix: Reuse What You Verify through Local Exact-Feature Correction for Faster Cached Diffusion Inference
- Predict to Skip: Linear Multistep Feature Forecasting for Efficient Diffusion Transformers
- Towards Memory-Efficient Autoregressive Video Generation via Instance-Specific Parametric Absorption
- LayerRecall: A State-Conditioned Memory Router for Long-Horizon Consistency in Video Generation
Source: arXiv cs.CV | 2026-09-01