Attention test seems uncompatible with latest xformers

[Jovp]

diffusers/src/diffusers/models/attention.py

Line 244 in 764b624

_ = xops.memory_efficient_attention(q, q, q)

It seems that the latest xformers has the memory_efficient_attention function under xops.ops and not xops in this case.

xops.version
'0.0.32.post2'

Maybe "import xformers as xops" should be "import xformers.ops as xops" ?

For reference:
https://github.com/facebookresearch/xformers/blob/c159edc05ae5a0192ab0558e834b946155790371/xformers/ops/fmha/__init__.py#L186

[Jovp]

Doing the import as suggested above seems to fix the problem. I am testing if xformers is working correctly now.

[ethanweber]

Hey all, I'm running into this now too. Should we make a PR for this?