only load state dict when the checkpoint is not None (huggingface#16673)

Author: laurahanu

src/transformers/modeling_utils.py

@@ -1792,7 +1792,7 @@ def from_pretrained(cls, pretrained_model_name_or_path: Optional[Union[str, os.P
 
         # load pt weights early so that we know which dtype to init the model under
         if from_pt:
-            if not is_sharded:
+            if not is_sharded and state_dict is None:
                 # Time to load the checkpoint
                 state_dict = load_state_dict(resolved_archive_file)
             # set dtype to instantiate the model under: