LIFT and PLACE: A Simple, Stable, and Effective Knowledge Distillation Framework for Lightweight Diffusion Models
Researchers have developed a new knowledge distillation framework called LIFT and PLACE to create more efficient diffusion models. This method addresses the difficulty students have in mimicking complex teacher models by using a coarse-to-fine alignment strategy. Experiments show its effectiveness across various diffusion model types and tasks, even achieving a low FID score of 15.73 with a significantly compressed student model. AI
IMPACT Enables the creation of smaller, more efficient diffusion models without significant performance loss.