cifar10 example error

[qiuyang163]

run
bash scripts/prune_ddpm_cifar10.sh 0.3
then I got error：

I seems that q、k、v in_features is not equal to out_features

@VainF

[VainF]

Hi @qiuyang163, thanks for reaching out. The script works on my side.

      (attn): ModuleList()
      (upsample): Upsample(
        (conv): Conv2d(179, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      )
    )
  )
  (norm_out): GroupNorm(32, 96, eps=1e-06, affine=True)
  (conv_out): Conv2d(96, 3, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
#Params: 35.7463 M => 19.8512 M
#MACs: 6.0655 G => 3.3928 G
Saving pruned model as run/finetune_simple_v2/cifar10_ours_T=0.05.pth/logs/post_training/pruned_model.pth

My env: pytorch==1.13.1, torch_pruning==1.3.1.

BTW, the in_features of qkv does not need to be equal to the out_features.

[qiuyang163]

Hi @qiuyang163, thanks for reaching out. The script works on my side.

      (attn): ModuleList()
      (upsample): Upsample(
        (conv): Conv2d(179, 192, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      )
    )
  )
  (norm_out): GroupNorm(32, 96, eps=1e-06, affine=True)
  (conv_out): Conv2d(96, 3, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
)
#Params: 35.7463 M => 19.8512 M
#MACs: 6.0655 G => 3.3928 G
Saving pruned model as run/finetune_simple_v2/cifar10_ours_T=0.05.pth/logs/post_training/pruned_model.pth

My env: pytorch==1.13.1, torch_pruning==1.3.1

yes，I got it，I succed in torch==1.13.1. But if I update to torch2.0, it failed
The reason is when use torch 2.0, diffusers lib will use attn_processor2.0 instead of attn_processor
Tks for reply

[VainF]

Thank you, will fix it soon.