[Tests] Inconsistent outputs for common fast tests on MPS

[anton-l]

There are three (so far) pipelines that don't produce the same outputs between two consecutive runs on MPS, despite having a warmup step in the common fast tests:

diffusers/tests/test_pipelines_common.py

Lines 67 to 73 in cd91fc0

	if torch_device == "mps" and self.pipeline_class in (
	DanceDiffusionPipeline,
	CycleDiffusionPipeline,
	StableDiffusionImg2ImgPipeline,
	):
	# FIXME: inconsistent outputs on MPS
	return

Needs further investigation to avoid skipping these tests

[patrickvonplaten]

Interesting! Thanks for flagging this here! I think especially StableDiffusionImg2ImgPipeline should give identical results. I wonder whether this is because of the stochastic sampling operation when encoding the image for img2img

[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]

I think currently we cannot do much as this seems to be a problem in native PyTorch/MPS no @pcuenca ?

[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.