ci: mark model_parallel tests as cuda specific (#35269)

`parallelize()` API is deprecated in favor of accelerate's `device_map="auto"`
and therefore is not accepting new features. At the same time `parallelize()`
implementation is currently CUDA-specific. This commit marks respective
ci tests with `@require_torch_gpu`.

Fixes: #35252

Signed-off-by: Dmitry Rogozhkin <[email protected]>

Author: dvrogozh

tests/test_modeling_common.py

@@ -3061,6 +3061,7 @@ def test_multi_gpu_data_parallel_forward(self):
             with torch.no_grad():
                 _ = model(**self._prepare_for_class(inputs_dict, model_class))
 
+    @require_torch_gpu
     @require_torch_multi_gpu
     def test_model_parallelization(self):
         if not self.test_model_parallel:
@@ -3123,6 +3124,7 @@ def get_current_gpu_memory_use():
             gc.collect()
             torch.cuda.empty_cache()
 
+    @require_torch_gpu
     @require_torch_multi_gpu
     def test_model_parallel_equal_results(self):
         if not self.test_model_parallel: