Correct `special_tokens_mask` when `add_special_tokens=False` (#2965)

Don't know of a use case where that would be useful, but this is more consistent

src/transformers/tokenization_utils.py

@@ -1233,7 +1233,10 @@ class PreTrainedTokenizer(object):
             token_type_ids = [0] * len(ids) + ([1] * len(pair_ids) if pair else [])
 
         if return_special_tokens_mask:
-            encoded_inputs["special_tokens_mask"] = self.get_special_tokens_mask(ids, pair_ids)
+            if add_special_tokens:
+                encoded_inputs["special_tokens_mask"] = self.get_special_tokens_mask(ids, pair_ids)
+            else:
+                encoded_inputs["special_tokens_mask"] = [0] * len(sequence)
 
         encoded_inputs["input_ids"] = sequence
         if return_token_type_ids: