SoftCache introduces an adaptive step size caching mechanism for diffusion transformers, replacing hard binary caching with a soft approach that uses a headroom signal to determine how far cached steps advance. By allowing steps to jump beyond scheduled nodes and shrink as a refresh nears, SoftCache achieves significant speedups—between 3.5× and 6.9×—across image and video models such as FLUX, HunyuanVideo, and Wan2.1, outperforming all baselines on four fidelity metrics.
arXiv:2602.24208v2 Announce Type: replace-cross Abstract: Diffusion models achieve state-of-the-art video generation quality, but their inference remains expensive due to the large number of sequenti...
arXiv:2607. 29398v1 Announce Type: new Abstract: Diffusion models have revolutionized generative tasks but incur high latency due to iterative denoising.
arXiv:2607. 23159v1 Announce Type: new Abstract: Test-time search lets small video diffusion models rival larger ones, but costs 2-10x more.
arXiv:2610.03577v1 Announce Type: cross Abstract: Diffusion Transformers (DiTs) can generate high-quality images and videos, but generating each sample requires multiple costly DiT forward passes. Tw...
arXiv:2608.29264v1 Announce Type: new Abstract: Diffusion-based visual generative models deliver strong image and video synthesis quality but incur high inference costs because sequential samplers re...
arXiv:2606. 31026v1 Announce Type: cross Abstract: We propose OTCache, a training-free framework for accelerating diffusion sampling via caching schedule prediction.
The paper investigates how block‑diffusion language models can use a constant‑size cache to enable efficient parallel decoding. By employing sequence mixers that summarize completed blocks into a reusable state and a block‑causal training objective, the authors pretrain three 3B block‑diffusion denoisers (attention, Mamba, and hybrid) on 300 B tokens. The resulting state‑space cache remains O(1) in memory and latency regardless of context length, yielding significant speed‑up and memory savings compared to traditional attention‑based caches, especially at very long sequences.
arXiv:2608. 13043v1 Announce Type: new Abstract: Diffusion models have achieved dominant performance in visual generation but suffer from substantial inference overhead.
arXiv:2609.32540v2 Announce Type: replace-cross Abstract: Few-step autoregressive video diffusion generates a long video by splitting the video into temporal chunks and generating chunk-by-chunk, eac...
arXiv:2610.10457v1 Announce Type: new Abstract: Diffusion Transformers (DiTs) achieve remarkable performance in video synthesis, but their iterative denoising process suffers from high inference late...
DIFFLET is a serving engine for diffusion transformer models on AWS Trainium that allows five model pipelines to share a single compilation backend. It splits joint attention across cores by attention head and uses a host controller to decide whether to run the denoiser or reuse previous outputs at each denoising step. Across five image and video model families, DIFFLET’s fused attention accelerates denoiser steps by 1.2–4.3× compared to PyTorch/XLA, and resident serving is 1.5–3.4× faster than standalone runs that reload weights.
arXiv:2609.39343v1 Announce Type: new Abstract: Diffusion caching accelerates generation by replacing transformer computation with cached or predicted features at selected denoising steps. We introdu...