search

huggingface / diffusers

🤗 Diffusers: State-of-the-art diffusion models for image and audio generation in PyTorch and FLAX.
https://huggingface.co/docs/diffusers
Apache License 2.0
25.61k stars 5.29k forks source link

Implementing EDICT for memory efficient backpropagation through chain/time/steps #5348

Closed ysig closed 11 months ago

ysig commented 1 year ago

Currently backpropagation through chain is impossible in diffusers, due to obvious memory limitations. However recently through invertible networks and MemCNN a recent technique known as EDICT [paper,code] has been proposed that allows to do this and it has been used successfully deployed in this diffusion editing method known as DOODL [paper, code]. As the code is already in diffusers, it would be cool if the team implemented this in the most efficient way package wise.

DN6 commented 1 year ago

@ysig Feel free to open a PR with a reference implementation. Maybe add it to the example training scripts?

github-actions[bot] commented 11 months ago

This issue has been automatically marked as stale because it has not had recent activity. If you think this still needs to be addressed please comment on this thread.

Please note that issues that do not follow the contributing guidelines are likely to be ignored.