set default submesh_tp_size to prevent unset local variable error (#3687)

* set default submesh_tp_size to prevent unset local variable error

* Apply style fixes

---------

Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>

Author: winglian

src/accelerate/accelerator.py

@@ -329,7 +329,10 @@ def __init__(
 
         if deepspeed_plugins is None:
             # First check if we're creating another `Accelerator` w/o setting `deepspeed_plugin`
-            if AcceleratorState._shared_state != {} and AcceleratorState().distributed_type == DistributedType.DEEPSPEED:
+            if (
+                AcceleratorState._shared_state != {}
+                and AcceleratorState().distributed_type == DistributedType.DEEPSPEED
+            ):
                 deepspeed_plugins = AcceleratorState().deepspeed_plugins
             else:
                 # init from env variables

src/accelerate/data_loader.py

@@ -1113,6 +1113,7 @@ def prepare_data_loader(
             # Given a device mesh (dp, tp) = (2, 3):
             # - From the data parallel perspective, ranks should be structured as: 0 0 0 1 1 1
             # - Processes with the same DP rank will receive the same batch.
+            submesh_tp_size = 1
             if "tp" in torch_device_mesh.mesh_dim_names:
                 submesh_tp_size = torch_device_mesh["tp"].size()
             process_index = process_index // submesh_tp_size