Reproduce “TamedFlow: Trajectory-Aware Distillation for Teacher-Assisted Diffusion Sampling” 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
Shuaiqi Cheng, Jiahao Zhang, Boyuan Gu, Yuxi Chen, Haiying Xu, Zexin Lu, Huiyu Zhou, Xuming Hu
TamedFlow trains a compact diffusion model to draft several denoising steps, then uses a learned router to accept drafts or roll them back for a step from the original teacher. On class-conditional ImageNet at 256×256, it distills a 675M-parameter teacher into a 130M-parameter student, reaching FID-50K 2.58 at 2.25× measured speedup. The teacher remains deployed, and results cover evaluated settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.