Moving certain pipelines slow tests to nightly (#4469)

* move audioldm tests to nightly

* move kandinsky im2img ddpm test to nightly

* move flax dpm test to nightly

* move diffedit dpm test to nightly

* move fp16 slow tests to nightly

Author: DN6

tests/pipelines/audioldm/test_audioldm.py

@@ -36,7 +36,7 @@
     PNDMScheduler,
     UNet2DConditionModel,
 )
-from diffusers.utils import is_xformers_available, slow, torch_device
+from diffusers.utils import is_xformers_available, nightly, slow, torch_device
 from diffusers.utils.testing_utils import enable_full_determinism
 
 from ..pipeline_params import TEXT_TO_AUDIO_BATCH_PARAMS, TEXT_TO_AUDIO_PARAMS
@@ -408,6 +408,27 @@ def test_audioldm(self):
         max_diff = np.abs(expected_slice - audio_slice).max()
         assert max_diff < 1e-2
 
+
+@nightly
+class AudioLDMPipelineNightlyTests(unittest.TestCase):
+    def tearDown(self):
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def get_inputs(self, device, generator_device="cpu", dtype=torch.float32, seed=0):
+        generator = torch.Generator(device=generator_device).manual_seed(seed)
+        latents = np.random.RandomState(seed).standard_normal((1, 8, 128, 16))
+        latents = torch.from_numpy(latents).to(device=device, dtype=dtype)
+        inputs = {
+            "prompt": "A hammer hitting a wooden surface",
+            "latents": latents,
+            "generator": generator,
+            "num_inference_steps": 3,
+            "guidance_scale": 2.5,
+        }
+        return inputs
+
     def test_audioldm_lms(self):
         audioldm_pipe = AudioLDMPipeline.from_pretrained("cvssp/audioldm")
         audioldm_pipe.scheduler = LMSDiscreteScheduler.from_config(audioldm_pipe.scheduler.config)

tests/pipelines/kandinsky/test_kandinsky_img2img.py

@@ -31,7 +31,7 @@
     VQModel,
 )
 from diffusers.pipelines.kandinsky.text_encoder import MCLIPConfig, MultilingualCLIP
-from diffusers.utils import floats_tensor, load_image, load_numpy, slow, torch_device
+from diffusers.utils import floats_tensor, load_image, load_numpy, nightly, slow, torch_device
 from diffusers.utils.testing_utils import enable_full_determinism, require_torch_gpu
 
 from ..test_pipelines_common import PipelineTesterMixin, assert_mean_pixel_difference
@@ -344,6 +344,16 @@ def test_kandinsky_img2img(self):
 
         assert_mean_pixel_difference(image, expected_image)
 
+
+@nightly
+@require_torch_gpu
+class KandinskyImg2ImgPipelineNightlyTests(unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
     def test_kandinsky_img2img_ddpm(self):
         expected_image = load_numpy(
             "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main"

tests/pipelines/stable_diffusion_2/test_stable_diffusion_diffedit.py

@@ -32,7 +32,7 @@
     StableDiffusionDiffEditPipeline,
     UNet2DConditionModel,
 )
-from diffusers.utils import load_image, slow
+from diffusers.utils import load_image, nightly, slow
 from diffusers.utils.testing_utils import enable_full_determinism, floats_tensor, require_torch_gpu, torch_device
 
 from ..pipeline_params import TEXT_GUIDED_IMAGE_INPAINTING_BATCH_PARAMS, TEXT_GUIDED_IMAGE_INPAINTING_PARAMS
@@ -347,6 +347,25 @@ def test_stable_diffusion_diffedit_full(self):
         )
         assert np.abs((expected_image - image).max()) < 5e-1
 
+
+@nightly
+@require_torch_gpu
+class StableDiffusionDiffEditPipelineNightlyTests(unittest.TestCase):
+    def tearDown(self):
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    @classmethod
+    def setUpClass(cls):
+        raw_image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/diffedit/fruit.png"
+        )
+
+        raw_image = raw_image.convert("RGB").resize((768, 768))
+
+        cls.raw_image = raw_image
+
     def test_stable_diffusion_diffedit_dpm(self):
         generator = torch.manual_seed(0)
 

tests/pipelines/stable_diffusion_2/test_stable_diffusion_flax.py

@@ -17,7 +17,7 @@
 import unittest
 
 from diffusers import FlaxDPMSolverMultistepScheduler, FlaxStableDiffusionPipeline
-from diffusers.utils import is_flax_available, slow
+from diffusers.utils import is_flax_available, nightly, slow
 from diffusers.utils.testing_utils import require_flax
 
 
@@ -65,6 +65,15 @@ def test_stable_diffusion_flax(self):
         print(f"output_slice: {output_slice}")
         assert jnp.abs(output_slice - expected_slice).max() < 1e-2
 
+
+@nightly
+@require_flax
+class FlaxStableDiffusion2PipelineNightlyTests(unittest.TestCase):
+    def tearDown(self):
+        # clean up the VRAM after each test
+        super().tearDown()
+        gc.collect()
+
     def test_stable_diffusion_dpm_flax(self):
         model_id = "stabilityai/stable-diffusion-2"
         scheduler, scheduler_params = FlaxDPMSolverMultistepScheduler.from_pretrained(model_id, subfolder="scheduler")

tests/pipelines/unidiffuser/test_unidiffuser.py

@@ -20,7 +20,7 @@
     UniDiffuserPipeline,
     UniDiffuserTextDecoder,
 )
-from diffusers.utils import floats_tensor, load_image, randn_tensor, slow, torch_device
+from diffusers.utils import floats_tensor, load_image, nightly, randn_tensor, slow, torch_device
 from diffusers.utils.testing_utils import require_torch_gpu
 
 from ..pipeline_params import TEXT_GUIDED_IMAGE_VARIATION_BATCH_PARAMS, TEXT_GUIDED_IMAGE_VARIATION_PARAMS
@@ -619,6 +619,56 @@ def test_unidiffuser_default_img2text_v1(self):
         expected_text_prefix = "An astronaut"
         assert text[0][: len(expected_text_prefix)] == expected_text_prefix
 
+
+@nightly
+@require_torch_gpu
+class UniDiffuserPipelineNightlyTests(unittest.TestCase):
+    def tearDown(self):
+        super().tearDown()
+        gc.collect()
+        torch.cuda.empty_cache()
+
+    def get_inputs(self, device, seed=0, generate_latents=False):
+        generator = torch.manual_seed(seed)
+        image = load_image(
+            "https://huggingface.co/datasets/hf-internal-testing/diffusers-images/resolve/main/unidiffuser/unidiffuser_example_image.jpg"
+        )
+        inputs = {
+            "prompt": "an elephant under the sea",
+            "image": image,
+            "generator": generator,
+            "num_inference_steps": 3,
+            "guidance_scale": 8.0,
+            "output_type": "numpy",
+        }
+        if generate_latents:
+            latents = self.get_fixed_latents(device, seed=seed)
+            for latent_name, latent_tensor in latents.items():
+                inputs[latent_name] = latent_tensor
+        return inputs
+
+    def get_fixed_latents(self, device, seed=0):
+        if type(device) == str:
+            device = torch.device(device)
+        latent_device = torch.device("cpu")
+        generator = torch.Generator(device=latent_device).manual_seed(seed)
+        # Hardcode the shapes for now.
+        prompt_latents = randn_tensor((1, 77, 768), generator=generator, device=device, dtype=torch.float32)
+        vae_latents = randn_tensor((1, 4, 64, 64), generator=generator, device=device, dtype=torch.float32)
+        clip_latents = randn_tensor((1, 1, 512), generator=generator, device=device, dtype=torch.float32)
+
+        # Move latents onto desired device.
+        prompt_latents = prompt_latents.to(device)
+        vae_latents = vae_latents.to(device)
+        clip_latents = clip_latents.to(device)
+
+        latents = {
+            "prompt_latents": prompt_latents,
+            "vae_latents": vae_latents,
+            "clip_latents": clip_latents,
+        }
+        return latents
+
     def test_unidiffuser_default_joint_v1_fp16(self):
         pipe = UniDiffuserPipeline.from_pretrained("thu-ml/unidiffuser-v1", torch_dtype=torch.float16)
         pipe.to(torch_device)