diff --git a/src/transformers/models/lxmert/modeling_lxmert.py b/src/transformers/models/lxmert/modeling_lxmert.py
index 1256c98b5de5d6..8d22daafa7dd8b 100644
--- a/src/transformers/models/lxmert/modeling_lxmert.py
+++ b/src/transformers/models/lxmert/modeling_lxmert.py
@@ -705,7 +705,7 @@ def __init__(self, config, lxmert_model_embedding_weights):
         self.decoder.weight = lxmert_model_embedding_weights
         self.bias = nn.Parameter(torch.zeros(lxmert_model_embedding_weights.size(0)))
 
-    def maybe_resize_bias(self, new_size: int):
+    def _maybe_resize_bias(self, new_size: int):
         if new_size != self.bias.shape[0]:
             self.bias.data = nn.functional.pad(self.bias.data, (0, new_size - self.bias.shape[0]), "constant", 0)
 
@@ -1082,7 +1082,7 @@ def __init__(self, config):
 
     def _tie_weights(self):
         self.cls.predictions.decoder.weight = self.lxmert.embeddings.word_embeddings.weight
-        self.cls.predictions.maybe_resize_bias(self.lxmert.embeddings.word_embeddings.weight.shape[0])
+        self.cls.predictions._maybe_resize_bias(self.lxmert.embeddings.word_embeddings.weight.shape[0])
 
     def resize_num_qa_labels(self, num_labels):
         """