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/test_modeling_llava.py

@@ -14,7 +14,6 @@
 # limitations under the License.
 """Testing suite for the PyTorch Llava model."""
 
-import gc
 import unittest
 
 import requests
@@ -28,6 +27,7 @@ from transformers import (
     is_vision_available,
 )
 from transformers.testing_utils import (
+    cleanup,
     require_bitsandbytes,
     require_torch,
     require_torch_gpu,
@@ -307,8 +307,7 @@ class LlavaForConditionalGenerationIntegrationTest(unittest.TestCase):
         self.processor = AutoProcessor.from_pretrained("llava-hf/bakLlava-v1-hf")
 
     def tearDown(self):
-        gc.collect()
-        torch.cuda.empty_cache()
+        cleanup(torch_device, gc_collect=True)
 
     @slow
     @require_bitsandbytes