hkz
I had the same issue :)
me too for 0.7.9 version.
> And another thing is to add some naive transformation as defenses(rotation, random croping, brightness change). It seems these naive methods are very effective. These defenses are effective when adding...
> @THUYimingLi I think I won't be able to add those spatial transformation methods recently... > > And I will not change the order of adding marks as #49 illustrates....
They add `transforms.Normalize(mean, std)` in `train_transform` instead of adding normalization layer to the model with nn.Sequential. You can find more details in https://github.com/Harry24k/adversarial-attacks-pytorch/issues/1.
same issue +1
If you are using DistributedDataParallel, try to convert BatchNorm layers to SyncBatchNorm ones.
那请问这个问题出现在哪呢？能否指条路？
> Does it now support gradient accumulation for multiple models? I think #1708 should fix it according to [comment](https://github.com/huggingface/diffusers/issues/4046#issuecomment-1686548737).