Fix clearing backend cache from device agnostic testing (#6075)

update

tests/models/test_models_prior.py

@@ -164,7 +164,7 @@ class PriorTransformerIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        backend_empty_cache()
+        backend_empty_cache(torch_device)
 
     @parameterized.expand(
         [

tests/models/test_models_unet_2d_condition.py

@@ -869,7 +869,7 @@ class UNet2DConditionModelIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        backend_empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_latents(self, seed=0, shape=(4, 4, 64, 64), fp16=False):
         dtype = torch.float16 if fp16 else torch.float32

tests/models/test_models_vae.py

@@ -485,7 +485,7 @@ class AutoencoderTinyIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        backend_empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_file_format(self, seed, shape):
         return f"gaussian_noise_s={seed}_shape={'_'.join([str(s) for s in shape])}.npy"
@@ -565,7 +565,7 @@ class AutoencoderKLIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        backend_empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_sd_image(self, seed=0, shape=(4, 3, 512, 512), fp16=False):
         dtype = torch.float16 if fp16 else torch.float32
@@ -820,7 +820,7 @@ class AsymmetricAutoencoderKLIntegrationTests(unittest.TestCase):
         # clean up the VRAM after each test
         super().tearDown()
         gc.collect()
-        backend_empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_sd_image(self, seed=0, shape=(4, 3, 512, 512), fp16=False):
         dtype = torch.float16 if fp16 else torch.float32

tests/pipelines/stable_diffusion_2/test_stable_diffusion.py

@@ -310,7 +310,7 @@ class StableDiffusion2PipelineSlowTests(unittest.TestCase):
     def tearDown(self):
         super().tearDown()
         gc.collect()
-        backend_empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_inputs(self, device, generator_device="cpu", dtype=torch.float32, seed=0):
         _generator_device = "cpu" if not generator_device.startswith("cuda") else "cuda"
@@ -531,7 +531,7 @@ class StableDiffusion2PipelineNightlyTests(unittest.TestCase):
     def tearDown(self):
         super().tearDown()
         gc.collect()
-        backend_empty_cache()
+        backend_empty_cache(torch_device)
 
     def get_inputs(self, device, generator_device="cpu", dtype=torch.float32, seed=0):
         _generator_device = "cpu" if not generator_device.startswith("cuda") else "cuda"