Use native memory efficient attention in PyTorch 2.0 if possible

[haotian-liu]

When users use PyTorch 2.0, and do not explicit enable memory efficient attention with xformers, this can potentially lead to OOM issues (while the user may believe that the efficient attention is automatically enabled with PyTorch 2.0).

[HuggingFaceDocBuilderDev]

The documentation is not available anymore as the PR was closed or merged.

[kashif]

@haotian-liu this class is to be deprecated and in fact, the PR to remove it is #2697

The attention_processor.Attention class is the recommended way which selects the appropriate Pytorch 2.0 processor if it is found:

https://github.com/huggingface/diffusers/blob/main/src/diffusers/models/attention_processor.py#L109

[github-actions]

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.

[patrickvonplaten]

Hey @haotian-liu ,

I think this has been fixed with: #3200

[haotian-liu]

Great, thank you! Closing this as fixed in #3200.