Fix hint in src/transformers/modeling_utils.py (#22074)

fix hint

src/transformers/modeling_utils.py

@@ -807,7 +807,7 @@ class ModuleUtilsMixin:
         return extended_attention_mask
 
     def get_extended_attention_mask(
-        self, attention_mask: Tensor, input_shape: Tuple[int], device: device = None, dtype: torch.float = None
+        self, attention_mask: Tensor, input_shape: Tuple[int], device: torch.device = None, dtype: torch.float = None
     ) -> Tensor:
         """
         Makes broadcastable attention and causal masks so that future and masked tokens are ignored.