TamedFlow: Trajectory-Aware Distillation\\for Teacher-Assisted Diffusion Sampling

Key points

• Two acceleration strategies are increasingly important: model compression makes each evaluation cheaper, while adaptive inference avoids selected teacher computation through reuse, forecasting, or speculative execution.1
• We propose TamedFlow, which trains a compressed DiT for this deployed role through cross-architecture feature alignment, a near-identity Velocity Alignment Head, and randomized rollout supervision.1
• At inference, a lightweight Learnable Trajectory Estimator routes completed drafts: accepted drafts bypass the teacher, while rejected drafts roll back and receive one teacher rescue step.1
• Distilling DiT-XL/2 into DiT-B/2, TamedFlow reaches FID $2.58$ at $2.25\times$ measured speedup over DDIM-50 on ImageNet $256\times256$.1

Sources (1)

• [1]TamedFlow: Trajectory-Aware Distillation\\for Teacher-Assisted Diffusion Sampling
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 19, 08:48 AM

  Diffusion Transformers (DiTs) deliver high-fidelity image and video generation, but inference requires many sequential model evaluations.

  Two acceleration strategies are increasingly important: model compression makes each evaluation cheaper, while adaptive inference avoids selected teacher computation through reuse, forecasting, or speculative execution.

Extractive summary: sentences quoted from the sources.