enable several pipeline integration tests on XPU (#11526)

* enable kandinsky2_2 integration test cases on XPU
Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* fix style
Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* enable latent_diffusion, dance_diffusion, musicldm, shap_e integration
uts on xpu
Signed-off-by: Yao Matrix <matrix.yao@intel.com>

* fix style
Signed-off-by: Yao Matrix <matrix.yao@intel.com>

---------
Signed-off-by: Yao Matrix <matrix.yao@intel.com>
Co-authored-by: Aryan <aryan@huggingface.co>

tests/pipelines/dance_diffusion/test_dance_diffusion.py

@@ -20,7 +20,14 @@ import numpy as np
 import torch
 
 from diffusers import DanceDiffusionPipeline, IPNDMScheduler, UNet1DModel
-from diffusers.utils.testing_utils import enable_full_determinism, nightly, require_torch_gpu, skip_mps, torch_device
+from diffusers.utils.testing_utils import (
+    backend_empty_cache,
+    enable_full_determinism,
+    nightly,
+    require_torch_accelerator,
+    skip_mps,
+    torch_device,
+)
 
 from ..pipeline_params import UNCONDITIONAL_AUDIO_GENERATION_BATCH_PARAMS, UNCONDITIONAL_AUDIO_GENERATION_PARAMS
 from ..test_pipelines_common import PipelineTesterMixin
@@ -116,19 +123,19 @@ class DanceDiffusionPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class PipelineIntegrationTests(unittest.TestCase):
     def setUp(self):
         # clean up the VRAM before each test
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def test_dance_diffusion(self):
         device = torch_device

tests/pipelines/kandinsky2_2/test_kandinsky_controlnet.py

@@ -28,13 +28,15 @@ from diffusers import (
     VQModel,
 )
 from diffusers.utils.testing_utils import (
+    backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
     load_image,
     load_numpy,
     nightly,
     numpy_cosine_similarity_distance,
-    require_torch_gpu,
+    require_torch_accelerator,
+    torch_device,
 )
 
 from ..test_pipelines_common import PipelineTesterMixin
@@ -226,19 +228,19 @@ class KandinskyV22ControlnetPipelineFastTests(PipelineTesterMixin, unittest.Test
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class KandinskyV22ControlnetPipelineIntegrationTests(unittest.TestCase):
     def setUp(self):
         # clean up the VRAM before each test
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def test_kandinsky_controlnet(self):
         expected_image = load_numpy(

tests/pipelines/kandinsky2_2/test_kandinsky_controlnet_img2img.py

@@ -29,13 +29,15 @@ from diffusers import (
     VQModel,
 )
 from diffusers.utils.testing_utils import (
+    backend_empty_cache,
     enable_full_determinism,
     floats_tensor,
     load_image,
     load_numpy,
     nightly,
     numpy_cosine_similarity_distance,
-    require_torch_gpu,
+    require_torch_accelerator,
+    torch_device,
 )
 
 from ..test_pipelines_common import PipelineTesterMixin
@@ -233,19 +235,19 @@ class KandinskyV22ControlnetImg2ImgPipelineFastTests(PipelineTesterMixin, unitte
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class KandinskyV22ControlnetImg2ImgPipelineIntegrationTests(unittest.TestCase):
     def setUp(self):
         # clean up the VRAM before each test
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def test_kandinsky_controlnet_img2img(self):
         expected_image = load_numpy(
@@ -309,4 +311,4 @@ class KandinskyV22ControlnetImg2ImgPipelineIntegrationTests(unittest.TestCase):
         assert image.shape == (512, 512, 3)
 
         max_diff = numpy_cosine_similarity_distance(expected_image.flatten(), image.flatten())
-        assert max_diff < 1e-4
+        assert max_diff < 5e-4

tests/pipelines/latent_diffusion/test_latent_diffusion.py

@@ -22,10 +22,11 @@ from transformers import CLIPTextConfig, CLIPTextModel, CLIPTokenizer
 
 from diffusers import AutoencoderKL, DDIMScheduler, LDMTextToImagePipeline, UNet2DConditionModel
 from diffusers.utils.testing_utils import (
+    backend_empty_cache,
     enable_full_determinism,
     load_numpy,
     nightly,
-    require_torch_gpu,
+    require_torch_accelerator,
     torch_device,
 )
 
@@ -136,17 +137,17 @@ class LDMTextToImagePipelineFastTests(PipelineTesterMixin, unittest.TestCase):
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class LDMTextToImagePipelineSlowTests(unittest.TestCase):
     def setUp(self):
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_inputs(self, device, dtype=torch.float32, seed=0):
         generator = torch.manual_seed(seed)
@@ -177,17 +178,17 @@ class LDMTextToImagePipelineSlowTests(unittest.TestCase):
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class LDMTextToImagePipelineNightlyTests(unittest.TestCase):
     def setUp(self):
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_inputs(self, device, dtype=torch.float32, seed=0):
         generator = torch.manual_seed(seed)

tests/pipelines/musicldm/test_musicldm.py

@@ -39,7 +39,13 @@ from diffusers import (
     UNet2DConditionModel,
 )
 from diffusers.utils import is_xformers_available
-from diffusers.utils.testing_utils import enable_full_determinism, nightly, require_torch_gpu, torch_device
+from diffusers.utils.testing_utils import (
+    backend_empty_cache,
+    enable_full_determinism,
+    nightly,
+    require_torch_accelerator,
+    torch_device,
+)
 
 from ..pipeline_params import TEXT_TO_AUDIO_BATCH_PARAMS, TEXT_TO_AUDIO_PARAMS
 from ..test_pipelines_common import PipelineTesterMixin
@@ -408,17 +414,17 @@ class MusicLDMPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class MusicLDMPipelineNightlyTests(unittest.TestCase):
     def setUp(self):
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_inputs(self, device, generator_device="cpu", dtype=torch.float32, seed=0):
         generator = torch.Generator(device=generator_device).manual_seed(seed)

tests/pipelines/shap_e/test_shap_e.py

@@ -21,7 +21,13 @@ from transformers import CLIPTextConfig, CLIPTextModelWithProjection, CLIPTokeni
 
 from diffusers import HeunDiscreteScheduler, PriorTransformer, ShapEPipeline
 from diffusers.pipelines.shap_e import ShapERenderer
-from diffusers.utils.testing_utils import load_numpy, nightly, require_torch_gpu, torch_device
+from diffusers.utils.testing_utils import (
+    backend_empty_cache,
+    load_numpy,
+    nightly,
+    require_torch_accelerator,
+    torch_device,
+)
 
 from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
 
@@ -222,19 +228,19 @@ class ShapEPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class ShapEPipelineIntegrationTests(unittest.TestCase):
     def setUp(self):
         # clean up the VRAM before each test
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def test_shap_e(self):
         expected_image = load_numpy(

tests/pipelines/shap_e/test_shap_e_img2img.py

@@ -23,11 +23,12 @@ from transformers import CLIPImageProcessor, CLIPVisionConfig, CLIPVisionModel
 from diffusers import HeunDiscreteScheduler, PriorTransformer, ShapEImg2ImgPipeline
 from diffusers.pipelines.shap_e import ShapERenderer
 from diffusers.utils.testing_utils import (
+    backend_empty_cache,
     floats_tensor,
     load_image,
     load_numpy,
     nightly,
-    require_torch_gpu,
+    require_torch_accelerator,
     torch_device,
 )
 
@@ -250,19 +251,19 @@ class ShapEImg2ImgPipelineFastTests(PipelineTesterMixin, unittest.TestCase):
 
 
 @nightly
-@require_torch_gpu
+@require_torch_accelerator
 class ShapEImg2ImgPipelineIntegrationTests(unittest.TestCase):
     def setUp(self):
         # clean up the VRAM before each test
         super().setUp()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def tearDown(self):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        torch.cuda.empty_cache()
+        backend_empty_cache(torch_device)
 
     def test_shap_e_img2img(self):
         input_image = load_image(