make `pipelines` tests device-agnostic (part1) #9399

faaany · 2024-09-09T22:02:31Z

What does this PR do?

Below are some evidences:

PASSED tests/pipelines/animatediff/test_animatediff.py::AnimateDiffPipelineFastTests::test_to_device
PASSED tests/pipelines/amused/test_amused.py::AmusedPipelineFastTests::test_model_cpu_offload_forward_pass
PASSED tests/pipelines/amused/test_amused.py::AmusedPipelineFastTests::test_cpu_offload_
forward_pass_twice
PASSED tests/pipelines/amused/test_amused_inpaint.py::AmusedInpaintPipelineFastTests::te
st_to_device

@yiyixuxu

sayakpaul · 2024-09-10T01:34:33Z

Could you provide some details about the machine you used to run the tests changed in this PR?

faaany · 2024-09-10T01:36:32Z

Could you provide some details about the machine you used to run the tests changed in this PR?

yes, I am running on Intel(R) Data Center GPU Max 1550.

faaany · 2024-09-10T01:45:15Z

Could you provide some details about the machine you used to run the tests changed in this PR?

xpu is the common device name for intel gpu.

faaany · 2024-09-12T07:54:10Z

Hi @sayakpaul , any concern on this PR?

faaany · 2024-09-13T04:49:37Z

could you pls help retrigger the CI? Thx a lot!

faaany · 2024-09-18T08:40:51Z

Hi @yiyixuxu, could you let me know your thoughts on this PR?

faaany · 2024-09-20T07:24:06Z

Hi folks, 2 weeks already, any feedback? I am waiting to enable more cases on XPU... If this is not the right approach, pls let me know as well. Thanks a lot!

@sayakpaul @yiyixuxu

sayakpaul · 2024-09-20T07:38:05Z

Please be a little more respectful towards the maintainers' time. It will be reviewed soon.

HuggingFaceDocBuilderDev · 2024-09-20T07:44:08Z

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.

faaany · 2024-09-20T07:51:54Z

Please be a little more respectful towards the maintainers' time. It will be reviewed soon.

oh, if my expression sounds impolite, sorry for that. And thanks for letting me know!

No hurries, I am just a bit worried that I might not be on the right track. Thanks for your understanding.

yao-matrix · 2024-10-13T23:33:43Z

Please be a little more respectful towards the maintainers' time. It will be reviewed soon.

@sayakpaul , could you pls take a review on this? This PR and following PRs are part of efforts to integrate intel gpu into hugging face ecosystems and making CIs can run as expected, thx.

sayakpaul

Thanks!

yiyixuxu

I left some comments!

tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py

faaany · 2024-10-23T07:54:03Z

@yiyixuxu thanks for the comments. I went through all the changes again and updated the places when necessary. Pls have a review. Thx!

faaany · 2024-10-31T02:37:46Z

Hi @yiyixuxu , can we move forward with this PR?

faaany · 2024-11-06T06:02:29Z

Hi guys, any concern about this PR?

yiyixuxu · 2024-11-08T08:13:34Z

tests/pipelines/deepfloyd_if/test_if.py

@@ -58,7 +58,7 @@ def get_dummy_inputs(self, device, seed=0):
    def test_save_load_optional_components(self):
        self._test_save_load_optional_components()

-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
+    @require_non_cpu


float16 does not work with all accelerator, no?

okay, we have 2 options here:

we allow these tests only to run on cuda or xpu.

we allow them to run on all accelerators and those failed accelerators should either support it or skip these test.

I think the 2 option is better because float16 is a common data type and failing will make people seriously considering it rather than ignoring it.

but I am fine with both. just let me know what you would prefer. Thx!

@faaany Let's go with Option 1 please. I feel like with 2, you just end up with noise in the CI when running on accelerators that don't support FP16.

Suggested change

@require_non_cpu

@unittest.skipIf(torch_device not in ["cuda", "xpu"], reason="float16 requires CUDA or XPU")

@require_accelerator

DN6 · 2024-11-08T12:34:14Z

src/diffusers/utils/testing_utils.py

@@ -373,6 +373,14 @@ def require_note_seq(test_case):
    return unittest.skipUnless(is_note_seq_available(), "test requires note_seq")(test_case)


+def require_non_cpu(test_case):


Suggested change

def require_non_cpu(test_case):

def require_accelerator(test_case):

DN6 · 2024-11-08T12:34:57Z

tests/pipelines/animatediff/test_animatediff.py

@@ -272,7 +278,7 @@ def test_inference_batch_single_identical(
        max_diff = np.abs(to_np(output_batch[0][0]) - to_np(output[0][0])).max()
        assert max_diff < expected_max_diff

-    @unittest.skipIf(torch_device != "cuda", reason="CUDA and CPU are required to switch devices")
+    @require_non_cpu


Suggested change

@require_non_cpu

@require_accelerator

DN6 · 2024-11-08T12:35:56Z

tests/pipelines/animatediff/test_animatediff_controlnet.py

@@ -21,7 +21,7 @@
 from diffusers.models.attention import FreeNoiseTransformerBlock
 from diffusers.utils import logging
 from diffusers.utils.import_utils import is_xformers_available
-from diffusers.utils.testing_utils import torch_device
+from diffusers.utils.testing_utils import require_non_cpu, torch_device


Suggested change

from diffusers.utils.testing_utils import require_non_cpu, torch_device

from diffusers.utils.testing_utils import require_accelerator, torch_device

DN6 · 2024-11-08T12:36:09Z

tests/pipelines/animatediff/test_animatediff_controlnet.py

@@ -281,7 +281,7 @@ def test_inference_batch_single_identical(
        max_diff = np.abs(to_np(output_batch[0][0]) - to_np(output[0][0])).max()
        assert max_diff < expected_max_diff

-    @unittest.skipIf(torch_device != "cuda", reason="CUDA and CPU are required to switch devices")
+    @require_non_cpu


Suggested change

@require_non_cpu

@require_accelerator

DN6 · 2024-11-08T12:44:36Z

tests/pipelines/deepfloyd_if/test_if.py

@@ -58,7 +58,7 @@ def get_dummy_inputs(self, device, seed=0):
    def test_save_load_optional_components(self):
        self._test_save_load_optional_components()

-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
+    @require_non_cpu


@faaany Let's go with Option 1 please. I feel like with 2, you just end up with noise in the CI when running on accelerators that don't support FP16.

Suggested change

@require_non_cpu

@unittest.skipIf(torch_device not in ["cuda", "xpu"], reason="float16 requires CUDA or XPU")

@require_accelerator

faaany added 6 commits September 9, 2024 14:47

enable on xpu

275b6af

add 1 more

f422d55

add one more

84540d9

enable more

b94f751

add 1 more

fa508a4

add more

a04fedf

faaany marked this pull request as ready for review September 9, 2024 22:55

faaany added 8 commits September 9, 2024 16:20

enable 1

7442dc1

enable more cases

658789a

enable

06c77c7

enable

1bf9b0c

update comment

6296e88

one more

6d0b378

enable 1

0d9f975

add more cases

732ff3a

faaany added 2 commits September 9, 2024 22:15

enable xpu

3803d93

Merge branch 'main' into enable_xpu

1825fd5

Merge branch 'main' into enable_xpu

0666a53

faaany added 2 commits September 12, 2024 22:38

add one more caswe

e00bcca

add more cases

6513201

sayakpaul requested a review from yiyixuxu September 13, 2024 05:46

faaany added 4 commits September 12, 2024 23:01

add 1

3cb0f96

add more

314e76d

add more cases

65c95e9

add case

2882ee4

Merge branch 'main' into enable_xpu

563124e

Merge branch 'main' into enable_xpu

feb9fd2

Merge branch 'main' into enable_xpu

0e39938

faaany added 2 commits September 23, 2024 12:59

Merge branch 'main' into enable_xpu

58de480

Merge branch 'main' into enable_xpu

ac90301

Merge branch 'main' into enable_xpu

2d08678

sayakpaul approved these changes Oct 14, 2024

View reviewed changes

sayakpaul requested a review from a-r-r-o-w October 14, 2024 03:29

yiyixuxu reviewed Oct 14, 2024

View reviewed changes

tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py Show resolved Hide resolved

tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py Outdated Show resolved Hide resolved

faaany and others added 7 commits October 15, 2024 14:55

Merge branch 'huggingface:main' into enable_xpu

d1b81ec

Merge branch 'huggingface:main' into enable_xpu

da9c5c9

update decorator

34a0a7b

Merge branch 'huggingface:main' into enable_xpu

3e7490e

update

bad2a3e

update

f157e98

update

359754e

faaany added 2 commits October 31, 2024 10:37

Merge branch 'main' into enable_xpu

9052f83

Merge branch 'main' into enable_xpu

ed55b90

yiyixuxu requested a review from DN6 November 8, 2024 01:38

yiyixuxu reviewed Nov 8, 2024

View reviewed changes

DN6 reviewed Nov 8, 2024

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

make `pipelines` tests device-agnostic (part1) #9399

make `pipelines` tests device-agnostic (part1) #9399

faaany commented Sep 9, 2024 •

edited

Loading

sayakpaul commented Sep 10, 2024

faaany commented Sep 10, 2024

faaany commented Sep 10, 2024

faaany commented Sep 12, 2024

faaany commented Sep 13, 2024

faaany commented Sep 18, 2024

faaany commented Sep 20, 2024 •

edited

Loading

sayakpaul commented Sep 20, 2024

HuggingFaceDocBuilderDev commented Sep 20, 2024

faaany commented Sep 20, 2024

yao-matrix commented Oct 13, 2024

sayakpaul left a comment

yiyixuxu left a comment

faaany commented Oct 23, 2024

faaany commented Oct 31, 2024

faaany commented Nov 6, 2024

yiyixuxu Nov 8, 2024 •

edited

Loading

faaany Nov 8, 2024 •

edited

Loading

DN6 Nov 8, 2024

DN6 Nov 8, 2024

DN6 Nov 8, 2024

DN6 Nov 8, 2024

DN6 Nov 8, 2024

DN6 Nov 8, 2024

	@require_non_cpu
	@unittest.skipIf(torch_device not in ["cuda", "xpu"], reason="float16 requires CUDA or XPU")
	@require_accelerator

		@@ -373,6 +373,14 @@ def require_note_seq(test_case):
		return unittest.skipUnless(is_note_seq_available(), "test requires note_seq")(test_case)


		def require_non_cpu(test_case):

	def require_non_cpu(test_case):
	def require_accelerator(test_case):

	from diffusers.utils.testing_utils import require_non_cpu, torch_device
	from diffusers.utils.testing_utils import require_accelerator, torch_device

make pipelines tests device-agnostic (part1) #9399

Are you sure you want to change the base?

make pipelines tests device-agnostic (part1) #9399

Conversation

faaany commented Sep 9, 2024 • edited Loading

What does this PR do?

sayakpaul commented Sep 10, 2024

faaany commented Sep 10, 2024

faaany commented Sep 10, 2024

faaany commented Sep 12, 2024

faaany commented Sep 13, 2024

faaany commented Sep 18, 2024

faaany commented Sep 20, 2024 • edited Loading

sayakpaul commented Sep 20, 2024

HuggingFaceDocBuilderDev commented Sep 20, 2024

faaany commented Sep 20, 2024

yao-matrix commented Oct 13, 2024

sayakpaul left a comment

Choose a reason for hiding this comment

yiyixuxu left a comment

Choose a reason for hiding this comment

faaany commented Oct 23, 2024

faaany commented Oct 31, 2024

faaany commented Nov 6, 2024

yiyixuxu Nov 8, 2024 • edited Loading

Choose a reason for hiding this comment

faaany Nov 8, 2024 • edited Loading

Choose a reason for hiding this comment

DN6 Nov 8, 2024

Choose a reason for hiding this comment

DN6 Nov 8, 2024

Choose a reason for hiding this comment

DN6 Nov 8, 2024

Choose a reason for hiding this comment

DN6 Nov 8, 2024

Choose a reason for hiding this comment

DN6 Nov 8, 2024

Choose a reason for hiding this comment

DN6 Nov 8, 2024

Choose a reason for hiding this comment

make `pipelines` tests device-agnostic (part1) #9399

make `pipelines` tests device-agnostic (part1) #9399

faaany commented Sep 9, 2024 •

edited

Loading

faaany commented Sep 20, 2024 •

edited

Loading

yiyixuxu Nov 8, 2024 •

edited

Loading

faaany Nov 8, 2024 •

edited

Loading