跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.25296 · v1

CalibCache: Cache-to-Full Feature Dynamics Calibration for Diffusion Transformers

首次提交 2026-11-19 07:43
版本更新 2026-11-19 07:43
雷达观测 2026-11-20 10:50
arXiv ↗PDF ↗

摘要

Diffusion Transformers generate high-quality images, but iterative sampling remains computationally expensive. Feature caching reduces the number of full network evaluations, yet its approximation error can become more severe under short sampling schedules, where fewer full-compute steps and a coarser sampling grid can make history-based extrapolation less reliable. In this paper, we propose CalibCache, a lightweight offline calibration method that corrects systematic discrepancies between cached and full-compute feature dynamics. Specifically, CalibCache calibrates feature acceleration, defined as the second-order finite difference across sampling steps. Using only a single pair of full-compute and naive-cache trajectories at each setting, CalibCache fits scalar acceleration maps offline by centered least squares. These maps are then applied during sampling without additional full-network evaluations or online optimization. Across FLUX.1-dev and Qwen-Image on DrawBench and DiT-XL/2 on ImageNet-256, CalibCache consistently improves fidelity or generation quality over TaylorSeer under 20-step sampling at cache intervals N=3,4,5N=3,4,5. At N=5N=5, it reduces FLUX.1-dev LPIPS from 0.5630.563 to 0.4290.429 while achieving a 2.86×2.86\times denoising speedup over full-compute sampling, and lowers DiT-XL/2 FID from 16.3916.39 to 10.3710.37 relative to TaylorSeer.

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v1 · 2026-11-19 07:43 · 最新观测版本

CalibCache: Cache-to-Full Feature Dynamics Calibration for Diffusion Transformers

Diffusion Transformers generate high-quality images, but iterative sampling remains computationally expensive. Feature caching reduces the number of full network evaluations, yet its approximation error can become more severe under short sampling schedules, where fewer full-compute steps and a coarser sampling grid can make history-based extrapolation less reliable. In this paper, we propose CalibCache, a lightweight offline calibration method that corrects systematic discrepancies between cached and full-compute feature dynamics. Specifically, CalibCache calibrates feature acceleration, defined as the second-order finite difference across sampling steps. Using only a single pair of full-compute and naive-cache trajectories at each setting, CalibCache fits scalar acceleration maps offline by centered least squares. These maps are then applied during sampling without additional full-network evaluations or online optimization. Across FLUX.1-dev and Qwen-Image on DrawBench and DiT-XL/2 on ImageNet-256, CalibCache consistently improves fidelity or generation quality over TaylorSeer under 20-step sampling at cache intervals N=3,4,5N=3,4,5. At N=5N=5, it reduces FLUX.1-dev LPIPS from 0.5630.563 to 0.4290.429 while achieving a 2.86×2.86\times denoising speedup over full-compute sampling, and lowers DiT-XL/2 FID from 16.3916.39 to 10.3710.37 relative to TaylorSeer.

查看此版本解读 →查看此版本原文 ↗