Open Njasa2k opened 6 days ago
Does this method work with SD3 or Flux (both diffusion transformers)? If not, any thoughts on adapting it for transformer-based architectures? Curious how the pruning strategies would handle their text/image components.
As written right now, the method works with SD3, but doesn't handle the image component. We intend to extend this work to more models in the future.
Does this method work with SD3 or Flux (both diffusion transformers)? If not, any thoughts on adapting it for transformer-based architectures? Curious how the pruning strategies would handle their text/image components.