diff --git a/src/transformers/trainer.py b/src/transformers/trainer.py
index 7267d79b3c86c5..549fb0f85ec071 100755
--- a/src/transformers/trainer.py
+++ b/src/transformers/trainer.py
@@ -2753,6 +2753,7 @@ def save_model(self, output_dir: Optional[str] = None, _internal_call: bool = Fa
             or ShardedDDPOption.ZERO_DP_3 in self.args.sharded_ddp
             or self.fsdp is not None
         ):
+            os.makedirs(output_dir, exist_ok=True)
             state_dict = self.model.state_dict()
 
             if self.args.should_save: