Merge branch 'xpu_more3' into enable_xpu

faaany · faaany · commit bf5e0b2dc296 · 2024-09-13T19:52:50.000-07:00
diff --git a/tests/models/test_modeling_common.py b/tests/models/test_modeling_common.py
@@ -48,6 +48,7 @@
     require_torch_2,
     require_torch_accelerator_with_training,
     require_torch_gpu,
+    require_torch_accelerator,
     require_torch_multi_gpu,
     run_test_in_subprocess,
     torch_device,
@@ -405,7 +406,7 @@ def test_set_xformers_attn_processor_for_determinism(self):
         assert torch.allclose(output, output_3, atol=self.base_precision)
         assert torch.allclose(output_2, output_3, atol=self.base_precision)
 
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_set_attn_processor_for_determinism(self):
         if self.uses_custom_attn_processor:
             return
@@ -752,7 +753,7 @@ def test_deprecated_kwargs(self):
                 " from `_deprecated_kwargs = [<deprecated_argument>]`"
             )
 
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_cpu_offload(self):
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
@@ -782,7 +783,7 @@ def test_cpu_offload(self):
 
                 self.assertTrue(torch.allclose(base_output[0], new_output[0], atol=1e-5))
 
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_disk_offload_without_safetensors(self):
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
@@ -816,7 +817,7 @@ def test_disk_offload_without_safetensors(self):
 
             self.assertTrue(torch.allclose(base_output[0], new_output[0], atol=1e-5))
 
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_disk_offload_with_safetensors(self):
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
@@ -875,7 +876,7 @@ def test_model_parallelism(self):
 
                 self.assertTrue(torch.allclose(base_output[0], new_output[0], atol=1e-5))
 
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_sharded_checkpoints(self):
         torch.manual_seed(0)
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
@@ -907,7 +908,7 @@ def test_sharded_checkpoints(self):
 
             self.assertTrue(torch.allclose(base_output[0], new_output[0], atol=1e-5))
 
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_sharded_checkpoints_with_variant(self):
         torch.manual_seed(0)
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
@@ -944,7 +945,7 @@ def test_sharded_checkpoints_with_variant(self):
 
             self.assertTrue(torch.allclose(base_output[0], new_output[0], atol=1e-5))
 
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_sharded_checkpoints_device_map(self):
         config, inputs_dict = self.prepare_init_args_and_inputs_for_common()
         model = self.model_class(**config).eval()
diff --git a/tests/pipelines/amused/test_amused.py b/tests/pipelines/amused/test_amused.py
@@ -22,11 +22,10 @@
 from diffusers import AmusedPipeline, AmusedScheduler, UVit2DModel, VQModel
 from diffusers.utils.testing_utils import (
     enable_full_determinism,
-    require_torch_gpu,
+    require_torch_accelerator,
     slow,
     torch_device,
 )
-
 from ..pipeline_params import TEXT_TO_IMAGE_BATCH_PARAMS, TEXT_TO_IMAGE_PARAMS
 from ..test_pipelines_common import PipelineTesterMixin
 
@@ -129,7 +128,7 @@ def test_inference_batch_single_identical(self):
 
 
 @slow
-@require_torch_gpu
+@require_torch_accelerator
 class AmusedPipelineSlowTests(unittest.TestCase):
     def test_amused_256(self):
         pipe = AmusedPipeline.from_pretrained("amused/amused-256")
diff --git a/tests/pipelines/amused/test_amused_img2img.py b/tests/pipelines/amused/test_amused_img2img.py
@@ -23,11 +23,10 @@
 from diffusers.utils import load_image
 from diffusers.utils.testing_utils import (
     enable_full_determinism,
-    require_torch_gpu,
+    require_torch_accelerator,
     slow,
     torch_device,
 )
-
 from ..pipeline_params import TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS, TEXT_GUIDED_IMAGE_VARIATION_PARAMS
 from ..test_pipelines_common import PipelineTesterMixin
 
@@ -131,7 +130,7 @@ def test_inference_batch_single_identical(self):
 
 
 @slow
-@require_torch_gpu
+@require_torch_accelerator
 class AmusedImg2ImgPipelineSlowTests(unittest.TestCase):
     def test_amused_256(self):
         pipe = AmusedImg2ImgPipeline.from_pretrained("amused/amused-256")
diff --git a/tests/pipelines/amused/test_amused_inpaint.py b/tests/pipelines/amused/test_amused_inpaint.py
@@ -23,11 +23,10 @@
 from diffusers.utils import load_image
 from diffusers.utils.testing_utils import (
     enable_full_determinism,
-    require_torch_gpu,
+    require_torch_accelerator,
     slow,
     torch_device,
 )
-
 from ..pipeline_params import TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS, TEXT_GUIDED_IMAGE_INPAINTING_PARAMS
 from ..test_pipelines_common import PipelineTesterMixin
 
@@ -135,7 +134,7 @@ def test_inference_batch_single_identical(self):
 
 
 @slow
-@require_torch_gpu
+@require_torch_accelerator
 class AmusedInpaintPipelineSlowTests(unittest.TestCase):
     def test_amused_256(self):
         pipe = AmusedInpaintPipeline.from_pretrained("amused/amused-256")
diff --git a/tests/pipelines/animatediff/test_animatediff_controlnet.py b/tests/pipelines/animatediff/test_animatediff_controlnet.py
@@ -281,7 +281,6 @@ 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")
     def test_to_device(self):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
@@ -297,14 +296,14 @@ def test_to_device(self):
         output_cpu = pipe(**self.get_dummy_inputs("cpu"))[0]
         self.assertTrue(np.isnan(output_cpu).sum() == 0)
 
-        pipe.to("cuda")
+        pipe.to(torch_device)
         model_devices = [
             component.device.type for component in pipe.components.values() if hasattr(component, "device")
         ]
-        self.assertTrue(all(device == "cuda" for device in model_devices))
+        self.assertTrue(all(device == torch_device for device in model_devices))
 
-        output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
-        self.assertTrue(np.isnan(to_np(output_cuda)).sum() == 0)
+        output_device = pipe(**self.get_dummy_inputs(torch_device))[0]
+        self.assertTrue(np.isnan(to_np(output_device)).sum() == 0)
 
     def test_to_dtype(self):
         components = self.get_dummy_components()
diff --git a/tests/pipelines/animatediff/test_animatediff_sdxl.py b/tests/pipelines/animatediff/test_animatediff_sdxl.py
@@ -212,7 +212,6 @@ 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")
     def test_to_device(self):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
@@ -228,14 +227,14 @@ def test_to_device(self):
         output_cpu = pipe(**self.get_dummy_inputs("cpu"))[0]
         self.assertTrue(np.isnan(output_cpu).sum() == 0)
 
-        pipe.to("cuda")
+        pipe.to(torch_device)
         model_devices = [
             component.device.type for component in pipe.components.values() if hasattr(component, "device")
         ]
-        self.assertTrue(all(device == "cuda" for device in model_devices))
+        self.assertTrue(all(device == torch_device for device in model_devices))
 
-        output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
-        self.assertTrue(np.isnan(to_np(output_cuda)).sum() == 0)
+        output_device = pipe(**self.get_dummy_inputs(torch_device))[0]
+        self.assertTrue(np.isnan(to_np(output_device)).sum() == 0)
 
     def test_to_dtype(self):
         components = self.get_dummy_components()
diff --git a/tests/pipelines/animatediff/test_animatediff_sparsectrl.py b/tests/pipelines/animatediff/test_animatediff_sparsectrl.py
@@ -345,7 +345,6 @@ def test_inference_batch_single_identical_use_simplified_condition_embedding_tru
         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")
     def test_to_device(self):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
@@ -361,13 +360,13 @@ def test_to_device(self):
         output_cpu = pipe(**self.get_dummy_inputs("cpu"))[0]
         self.assertTrue(np.isnan(output_cpu).sum() == 0)
 
-        pipe.to("cuda")
+        pipe.to(torch_device)
         model_devices = [
             component.device.type for component in pipe.components.values() if hasattr(component, "device")
         ]
-        self.assertTrue(all(device == "cuda" for device in model_devices))
+        self.assertTrue(all(device == torch_device for device in model_devices))
 
-        output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
+        output_cuda = pipe(**self.get_dummy_inputs(torch_device))[0]
         self.assertTrue(np.isnan(to_np(output_cuda)).sum() == 0)
 
     def test_to_dtype(self):
diff --git a/tests/pipelines/animatediff/test_animatediff_video2video.py b/tests/pipelines/animatediff/test_animatediff_video2video.py
@@ -258,7 +258,6 @@ 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")
     def test_to_device(self):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
@@ -274,14 +273,14 @@ def test_to_device(self):
         output_cpu = pipe(**self.get_dummy_inputs("cpu"))[0]
         self.assertTrue(np.isnan(output_cpu).sum() == 0)
 
-        pipe.to("cuda")
+        pipe.to(torch_device)
         model_devices = [
             component.device.type for component in pipe.components.values() if hasattr(component, "device")
         ]
-        self.assertTrue(all(device == "cuda" for device in model_devices))
+        self.assertTrue(all(device == torch_device for device in model_devices))
 
-        output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
-        self.assertTrue(np.isnan(to_np(output_cuda)).sum() == 0)
+        output_device = pipe(**self.get_dummy_inputs(torch_device))[0]
+        self.assertTrue(np.isnan(to_np(output_device)).sum() == 0)
 
     def test_to_dtype(self):
         components = self.get_dummy_components()
diff --git a/tests/pipelines/controlnet_xs/test_controlnetxs.py b/tests/pipelines/controlnet_xs/test_controlnetxs.py
@@ -306,7 +306,6 @@ def test_multi_vae(self):
 
             assert out_vae_np.shape == out_np.shape
 
-    @unittest.skipIf(torch_device != "cuda", reason="CUDA and CPU are required to switch devices")
     def test_to_device(self):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
@@ -322,14 +321,14 @@ def test_to_device(self):
         output_cpu = pipe(**self.get_dummy_inputs("cpu"))[0]
         self.assertTrue(np.isnan(output_cpu).sum() == 0)
 
-        pipe.to("cuda")
+        pipe.to(torch_device)
         model_devices = [
             component.device.type for component in pipe.components.values() if hasattr(component, "device")
         ]
-        self.assertTrue(all(device == "cuda" for device in model_devices))
+        self.assertTrue(all(device == torch_device for device in model_devices))
 
-        output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
-        self.assertTrue(np.isnan(to_np(output_cuda)).sum() == 0)
+        output_device = pipe(**self.get_dummy_inputs(torch_device))[0]
+        self.assertTrue(np.isnan(to_np(output_device)).sum() == 0)
 
 
 @slow
diff --git a/tests/pipelines/deepfloyd_if/test_if.py b/tests/pipelines/deepfloyd_if/test_if.py
@@ -58,7 +58,6 @@ 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")
     def test_save_load_float16(self):
         # Due to non-determinism in save load of the hf-internal-testing/tiny-random-t5 text encoder
         super().test_save_load_float16(expected_max_diff=1e-1)
diff --git a/tests/pipelines/deepfloyd_if/test_if_img2img.py b/tests/pipelines/deepfloyd_if/test_if_img2img.py
@@ -70,12 +70,10 @@ def test_save_load_optional_components(self):
     def test_xformers_attention_forwardGenerator_pass(self):
         self._test_xformers_attention_forwardGenerator_pass(expected_max_diff=1e-3)
 
-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
     def test_save_load_float16(self):
         # Due to non-determinism in save load of the hf-internal-testing/tiny-random-t5 text encoder
         super().test_save_load_float16(expected_max_diff=1e-1)
 
-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
     def test_float16_inference(self):
         super().test_float16_inference(expected_max_diff=1e-1)
 
diff --git a/tests/pipelines/deepfloyd_if/test_if_img2img_superresolution.py b/tests/pipelines/deepfloyd_if/test_if_img2img_superresolution.py
@@ -72,7 +72,6 @@ def test_xformers_attention_forwardGenerator_pass(self):
     def test_save_load_optional_components(self):
         self._test_save_load_optional_components()
 
-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
     def test_save_load_float16(self):
         # Due to non-determinism in save load of the hf-internal-testing/tiny-random-t5 text encoder
         super().test_save_load_float16(expected_max_diff=1e-1)
diff --git a/tests/pipelines/deepfloyd_if/test_if_inpainting.py b/tests/pipelines/deepfloyd_if/test_if_inpainting.py
@@ -72,7 +72,6 @@ def test_xformers_attention_forwardGenerator_pass(self):
     def test_save_load_optional_components(self):
         self._test_save_load_optional_components()
 
-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
     def test_save_load_float16(self):
         # Due to non-determinism in save load of the hf-internal-testing/tiny-random-t5 text encoder
         super().test_save_load_float16(expected_max_diff=1e-1)
diff --git a/tests/pipelines/deepfloyd_if/test_if_inpainting_superresolution.py b/tests/pipelines/deepfloyd_if/test_if_inpainting_superresolution.py
@@ -74,7 +74,6 @@ def test_xformers_attention_forwardGenerator_pass(self):
     def test_save_load_optional_components(self):
         self._test_save_load_optional_components()
 
-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
     def test_save_load_float16(self):
         # Due to non-determinism in save load of the hf-internal-testing/tiny-random-t5 text encoder
         super().test_save_load_float16(expected_max_diff=1e-1)
diff --git a/tests/pipelines/deepfloyd_if/test_if_superresolution.py b/tests/pipelines/deepfloyd_if/test_if_superresolution.py
@@ -67,7 +67,6 @@ def test_xformers_attention_forwardGenerator_pass(self):
     def test_save_load_optional_components(self):
         self._test_save_load_optional_components()
 
-    @unittest.skipIf(torch_device != "cuda", reason="float16 requires CUDA")
     def test_save_load_float16(self):
         # Due to non-determinism in save load of the hf-internal-testing/tiny-random-t5 text encoder
         super().test_save_load_float16(expected_max_diff=1e-1)
diff --git a/tests/pipelines/pag/test_pag_animatediff.py b/tests/pipelines/pag/test_pag_animatediff.py
@@ -218,7 +218,6 @@ def test_dict_tuple_outputs_equivalent(self):
             expected_slice = np.array([0.5295, 0.3947, 0.5300, 0.4864, 0.4518, 0.5315, 0.5440, 0.4775, 0.5538])
         return super().test_dict_tuple_outputs_equivalent(expected_slice=expected_slice)
 
-    @unittest.skipIf(torch_device != "cuda", reason="CUDA and CPU are required to switch devices")
     def test_to_device(self):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
@@ -234,14 +233,14 @@ def test_to_device(self):
         output_cpu = pipe(**self.get_dummy_inputs("cpu"))[0]
         self.assertTrue(np.isnan(output_cpu).sum() == 0)
 
-        pipe.to("cuda")
+        pipe.to(torch_device)
         model_devices = [
             component.device.type for component in pipe.components.values() if hasattr(component, "device")
         ]
-        self.assertTrue(all(device == "cuda" for device in model_devices))
+        self.assertTrue(all(device == torch_device for device in model_devices))
 
-        output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
-        self.assertTrue(np.isnan(to_np(output_cuda)).sum() == 0)
+        output_device = pipe(**self.get_dummy_inputs(torch_device))[0]
+        self.assertTrue(np.isnan(to_np(output_device)).sum() == 0)
 
     def test_to_dtype(self):
         components = self.get_dummy_components()
diff --git a/tests/pipelines/pia/test_pia.py b/tests/pipelines/pia/test_pia.py
@@ -278,7 +278,6 @@ 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")
     def test_to_device(self):
         components = self.get_dummy_components()
         pipe = self.pipeline_class(**components)
@@ -294,14 +293,14 @@ def test_to_device(self):
         output_cpu = pipe(**self.get_dummy_inputs("cpu"))[0]
         self.assertTrue(np.isnan(output_cpu).sum() == 0)
 
-        pipe.to("cuda")
+        pipe.to(torch_device)
         model_devices = [
             component.device.type for component in pipe.components.values() if hasattr(component, "device")
         ]
-        self.assertTrue(all(device == "cuda" for device in model_devices))
+        self.assertTrue(all(device == torch_device for device in model_devices))
 
-        output_cuda = pipe(**self.get_dummy_inputs("cuda"))[0]
-        self.assertTrue(np.isnan(to_np(output_cuda)).sum() == 0)
+        output_device = pipe(**self.get_dummy_inputs(torch_device))[0]
+        self.assertTrue(np.isnan(to_np(output_device)).sum() == 0)
 
     def test_to_dtype(self):
         components = self.get_dummy_components()
diff --git a/tests/pipelines/semantic_stable_diffusion/test_semantic_diffusion.py b/tests/pipelines/semantic_stable_diffusion/test_semantic_diffusion.py
@@ -237,7 +237,6 @@ def test_semantic_diffusion_no_safety_checker(self):
         image = pipe("example prompt", num_inference_steps=2).images[0]
         assert image is not None
 
-    @unittest.skipIf(torch_device != "cuda", "This test requires a GPU")
     def test_semantic_diffusion_fp16(self):
         """Test that stable diffusion works with fp16"""
         unet = self.dummy_cond_unet
diff --git a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_attend_and_excite.py b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_attend_and_excite.py
diff --git a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_depth.py
diff --git a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_upscale.py b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_upscale.py
diff --git a/tests/pipelines/stable_diffusion_2/test_stable_diffusion_v_pred.py b/tests/pipelines/stable_diffusion_2/test_stable_diffusion_v_pred.py
diff --git a/tests/pipelines/stable_diffusion_safe/test_safe_diffusion.py b/tests/pipelines/stable_diffusion_safe/test_safe_diffusion.py
diff --git a/tests/pipelines/stable_video_diffusion/test_stable_video_diffusion.py b/tests/pipelines/stable_video_diffusion/test_stable_video_diffusion.py
diff --git a/tests/pipelines/text_to_video_synthesis/test_text_to_video_zero_sdxl.py b/tests/pipelines/text_to_video_synthesis/test_text_to_video_zero_sdxl.py
