Reproduce “Simplifying and Improving On-Policy Distillation for Few-Step Diffusion” 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
Jiawei Zhang, Ziyu Wan, Nanye Ma, Long Zhao, Dong Chen, Hongyu Liu, Qifeng Chen, Dongdong Chen
The paper presents SMD, a method for distilling text-to-image models into four-step students. The student learns from teacher guidance queried along its own coarse rollout; a training-only Pivot residual represents local variation, while a repulsion term separates its prediction from a reference. At 4 NFE, SMD exceeds the 50-NFE FLUX.1-dev teacher on GenEval and DPG-Bench, but not universally.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.