Fix failing sharded_checkpoint tests that fail when pytorch 1.13 is n…

…ot installed (#1988)

tests/trainer/test_sharded_checkpoint.py

@@ -90,8 +90,9 @@ def _compare_model_params_between_state_dicts(state_dict1, state_dict2):
 
 @pytest.mark.gpu
 @world_size(2)
+@pytest.mark.skipif(version.parse(torch.__version__) < version.parse('1.13.0'),
+                    reason='requires PyTorch 1.13 or higher')
 def test_fsdp_full_state_dict_save(world_size, tmp_path: pathlib.Path):
-
     save_folder = tmp_path
     save_filename = 'rank{rank}.pt'
     num_features = 3
@@ -176,8 +177,9 @@ def test_fsdp_full_state_dict_save(world_size, tmp_path: pathlib.Path):
 
 @pytest.mark.gpu
 @world_size(2)
+@pytest.mark.skipif(version.parse(torch.__version__) < version.parse('1.13.0'),
+                    reason='requires PyTorch 1.13 or higher')
 def test_fsdp_full_state_dict_load(world_size, tmp_path: pathlib.Path):
-
     save_folder = tmp_path
     save_filename = 'rank{rank}.pt'
     trainer1 = get_trainer(save_folder=str(save_folder), save_filename=save_filename, fsdp_state_dict_type='full')
@@ -197,10 +199,9 @@ def test_fsdp_full_state_dict_load(world_size, tmp_path: pathlib.Path):
 @pytest.mark.gpu
 @world_size(2)
 @pytest.mark.parametrize('state_dict_type', ['local', 'sharded'])
+@pytest.mark.skipif(version.parse(torch.__version__) < version.parse('1.13.0'),
+                    reason='requires PyTorch 1.13 or higher')
 def test_fsdp_partitioned_state_dict_save(world_size, tmp_path: pathlib.Path, state_dict_type: str):
-
-    if version.parse(torch.__version__) < version.parse('1.13.0'):
-        pytest.skip()
     pytest.importorskip('torch.distributed.fsdp.fully_sharded_data_parallel')
     from torch.distributed.fsdp.fully_sharded_data_parallel import ShardedTensor
     save_folder = tmp_path
@@ -304,8 +305,9 @@ def test_fsdp_partitioned_state_dict_save(world_size, tmp_path: pathlib.Path, st
 @pytest.mark.gpu
 @world_size(2)
 @pytest.mark.parametrize('state_dict_type', ['local', 'sharded'])
+@pytest.mark.skipif(version.parse(torch.__version__) < version.parse('1.13.0'),
+                    reason='requires PyTorch 1.13 or higher')
 def test_fsdp_partitioned_state_dict_load(world_size, tmp_path: pathlib.Path, state_dict_type: str):
-
     save_folder = tmp_path
     save_filename = 'rank{rank}.pt'
     trainer1 = get_trainer(save_folder=str(save_folder),