[Tests] Non-deterministic depth estimator in `StableDiffusiondepth2imgPipelineFastTests`

[anton-l]

DPTForDepthEstimation might be causing less reproducible results between two consecutive runs than other transformers models, which motivated overriding the tolerances of common fast tests, e.g. 3e-5 here:

diffusers/tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py

Lines 155 to 174 in cd91fc0

	def test_save_load_local(self):
	components = self.get_dummy_components()
	pipe = self.pipeline_class(**components)
	pipe.to(torch_device)
	pipe.set_progress_bar_config(disable=None)
	inputs = self.get_dummy_inputs(torch_device)
	output = pipe(**inputs)[0]
	with tempfile.TemporaryDirectory() as tmpdir:
	pipe.save_pretrained(tmpdir)
	pipe_loaded = self.pipeline_class.from_pretrained(tmpdir)
	pipe_loaded.to(torch_device)
	pipe_loaded.set_progress_bar_config(disable=None)
	inputs = self.get_dummy_inputs(torch_device)
	output_loaded = pipe_loaded(**inputs)[0]
	max_diff = np.abs(output - output_loaded).max()
	self.assertLess(max_diff, 3e-5)

Needs further investigation, the common fast tests should ideally be applicable to all pipelines.

[patrickvonplaten]

I noticed this too, I noticed however that the results stay within a 1e-4 range which is good enough IMO. Should be fixed by: #1669

[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]

1e-4 range is good enough for such complex models so I think we can close this one

[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.