[tests] make cuda-only tests device-agnostic (#35607)

* intial commit

* remove unrelated files

* further remove

* Update test_trainer.py

* fix style

tests/models/llama/test_modeling_llama.py

@@ -26,7 +26,6 @@ from transformers.testing_utils import (
     require_read_token,
     require_torch,
     require_torch_accelerator,
-    require_torch_gpu,
     slow,
     torch_device,
 )
@@ -541,7 +540,7 @@ class LlamaModelTest(ModelTesterMixin, GenerationTesterMixin, PipelineTesterMixi
             config = _reinitialize_config(base_config, {"rope_scaling": {"rope_type": "linear"}})  # missing "factor"
 
 
-@require_torch_gpu
+@require_torch_accelerator
 class LlamaIntegrationTest(unittest.TestCase):
     # This variable is used to determine which CUDA device are we using for our runners (A10 or T4)
     # Depending on the hardware we get different logits / generations
@@ -695,7 +694,7 @@ class LlamaIntegrationTest(unittest.TestCase):
         self.assertEqual(EXPECTED_TEXT_COMPLETION, text)
 
     @slow
-    @require_torch_gpu
+    @require_torch_accelerator
     @require_read_token
     def test_compile_static_cache(self):
         # `torch==2.2` will throw an error on this test (as in other compilation tests), but torch==2.1.2 and torch>2.2