fix optimizer file name

paddlenlp/trainer/plugins/unified_checkpoint.py

@@ -1942,7 +1942,7 @@ def get_sharded_file_name(args, file_name, is_optimizer=False):
     else:
         hcg = fleet.get_hybrid_communicate_group()
         dp_group = hcg.get_data_parallel_group()
-        size = args.world_size if args.use_expert_parallel else dp_group.nranks
+        size = dp_group.nranks if not args.use_expert_parallel else 1
         shard_file = file_name.replace(
             ".pdparams", f"-{args.logical_process_index + 1:05d}-of-{args.world_size//size:05d}.pdparams"
         )
@@ -2246,7 +2246,7 @@ def merge_tensor_parallel_with_shard(state_dict, tp_actions, all_filter_keys):
 
     if len(tp_actions) > 0:
         for x in tp_actions.keys():
-            logger.warning(f"key <{x}> need to merge tensor parallel but we can't find in model state.")
+            logger.debug(f"key <{x}> need to merge tensor parallel but we can't find in model state.")
 
     return state_dict_to_save