Reproduce “SWIFT: Self-guided Weak-to-strong dIFfusion models via Target extrapolation” 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

Yanjin He, Cuifeng Shen, Wanzhou Lei, Maohua Li, Tao Lan, Kan Liu, Hanlin Tang, Qing Qu

Swift changes diffusion training targets by asking a denoiser to predict beyond the clean sample, away from a weaker, frozen guide. The guide is discarded after training, leaving the original sampler unchanged. On ImageNet, a matched 615-million-parameter configuration lowers unguided FID-50K from 2.09 to 1.75 after 80 epochs; this result does not establish why gains occur.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.