diffusionmodules.make_attn: fall back to vanilla if xformers is not available

[akx]

The default YAML file for SDXL specifies vanilla-xformers, which won't work if xformers is not available. This adds fallback logic like that in

generative-models/sgm/modules/attention.py

Lines 395 to 400 in e5dc966

	if attn_mode != "softmax" and not XFORMERS_IS_AVAILABLE:
	print(
	f"Attention mode '{attn_mode}' is not available. Falling back to native attention. "
	f"This is not a problem in Pytorch >= 2.0. FYI, you are running with PyTorch version {torch.__version__}"
	)
	attn_mode = "softmax"

[akx]

@benjaminaubin For what it's worth, I tested that this works on my mac (in a future branch that stacks this, #49, and other things so using mps is possible).

[benjaminaubin]

@benjaminaubin For what it's worth, I tested that this works on my mac (in a future branch that stacks this, #49, and other things so using mps is possible).

Yes my bad your condition is valid if you use torch >= 2.0

[benjaminaubin]

@akx please resolve conflicts with the previous merged branch

[akx]

@benjaminaubin rebased