avoid calling gc.collect and cuda.empty_cache (#34514)

* update

* update

* update

* update

* update

---------

Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>

tests/models/llava_onevision/test_modeling_llava_onevision.py

@@ -14,7 +14,6 @@
 # limitations under the License.
 """Testing suite for the PyTorch Llava-NeXT model."""
 
-import gc
 import unittest
 
 import numpy as np
@@ -29,6 +28,7 @@ from transformers import (
     is_vision_available,
 )
 from transformers.testing_utils import (
+    cleanup,
     require_bitsandbytes,
     require_torch,
     slow,
@@ -336,8 +336,7 @@ class LlavaOnevisionForConditionalGenerationIntegrationTest(unittest.TestCase):
         self.prompt_video = "user\n<video>\nWhat do you see in this video?<|im_end|>\n<|im_start|>assistant\n"
 
     def tearDown(self):
-        gc.collect()
-        torch.cuda.empty_cache()
+        cleanup(torch_device, gc_collect=True)
 
     @slow
     @require_bitsandbytes