Reproduce “Z-Cache: Accelerating Diffusion Transformers via Self-Reflection” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Zegang Cheng, Zhikai Wang, Jiacheng Liu, Xiaobing Tu, Chang Zou, Junjie Chen, Yue Ma, Zhiyuan Ma and 3 more
Z-Cache revises cached features by replaying a provisional window after the next scheduled computation provides a future anchor. It blends predictions from past and future anchors, then continues from the replayed latent. In an N=8 FLUX.1-dev DrawBench test, it improved PSNR, SSIM, and LPIPS over TaylorSeer at 6.22× F-Speed, but elapsed time increased.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.