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

* intial commit

* remove unrelated files

* further remove

* Update test_trainer.py

* fix style

tests/generation/test_utils.py

@@ -33,6 +33,7 @@ from transformers.testing_utils import (
     require_flash_attn,
     require_optimum_quanto,
     require_torch,
+    require_torch_accelerator,
     require_torch_gpu,
     require_torch_multi_accelerator,
     require_torch_multi_gpu,
@@ -2043,7 +2044,7 @@ class GenerationTesterMixin:
                 model.generate(**generation_kwargs, **inputs_dict)
 
     @pytest.mark.generate
-    @require_torch_gpu
+    @require_torch_accelerator
     @slow
     def test_generate_compile_model_forward(self):
         """
@@ -3791,10 +3792,12 @@ class GenerationIntegrationTests(unittest.TestCase, GenerationIntegrationTestsMi
         self.assertTrue(input_length <= out.shape[-1] <= input_length + 20)
 
     @slow
-    @require_torch_gpu
+    @require_torch_accelerator
     def test_assisted_decoding_model_in_gpu_assistant_in_cpu(self):
         # PT-only test: TF doesn't support assisted decoding yet.
-        model = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-MistralForCausalLM").to("cuda")
+        model = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-MistralForCausalLM").to(
+            torch_device
+        )
         assistant = AutoModelForCausalLM.from_pretrained("hf-internal-testing/tiny-random-MistralForCausalLM").to(
             "cpu"
         )