Fix token_type_ids for XLM-R

src/transformers/tokenization_xlm_roberta.py

@@ -176,10 +176,7 @@ class XLMRobertaTokenizer(PreTrainedTokenizer):
     def create_token_type_ids_from_sequences(self, token_ids_0, token_ids_1=None):
         """
         Creates a mask from the two sequences passed to be used in a sequence-pair classification task.
-        A RoBERTa sequence pair mask has the following format:
-        0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 1
-        | first sequence    | second sequence
-
+        RoBERTa does not make use of token type ids, therefore a list of zeros is returned.
         if token_ids_1 is None, only returns the first portion of the mask (0's).
         """
         sep = [self.sep_token_id]
@@ -187,7 +184,7 @@ class XLMRobertaTokenizer(PreTrainedTokenizer):
 
         if token_ids_1 is None:
             return len(cls + token_ids_0 + sep) * [0]
-        return len(cls + token_ids_0 + sep + sep) * [0] + len(token_ids_1 + sep) * [1]
+        return len(cls + token_ids_0 + sep + sep + token_ids_1 + sep) * [0]
 
     @property
     def vocab_size(self):