Does this work on Stable Diffusion 2.1 from the diffusers library?

[johnrachwan123]

No description provided.

[VainF]

Not yet. But we plan to support Stable Diffusion in the following months. This repo currently works for Latent Diffusion Models (Pruning-only) and DDPMs (Pruning + Training)

[Shan-handsome]

Thanks for the author's fantastic work.
I ran the code for LDM and found a runtime error:
"RuntimeError: CUDA error: device-side assert triggered"
It occurred at group_norm calculation (local_norm = local_norm[idxs])
Does there anyone have the same issue?

BTW, the script I ran is:
python ldm_prune.py
--model_path CompVis/ldm-celebahq-256
--save_path run/pruned/ldm_celeba_pruned
--pruning_ratio 0.05
--pruner magnitude
--device cuda:0
--batch_size 4

[VainF]

Thanks for the author's fantastic work. I ran the code for LDM and found a runtime error: "RuntimeError: CUDA error: device-side assert triggered" It occurred at group_norm calculation (local_norm = local_norm[idxs]) Does there anyone have the same issue?

BTW, the script I ran is: python ldm_prune.py --model_path CompVis/ldm-celebahq-256 --save_path run/pruned/ldm_celeba_pruned --pruning_ratio 0.05 --pruner magnitude --device cuda:0 --batch_size 4

Hi @Shan-handsome, thanks for your feedback. We will check it ASAP.