Fix misleading RoBERTa token type ids

src/transformers/tokenization_roberta.py

@@ -144,9 +144,7 @@ class RobertaTokenizer(GPT2Tokenizer):
     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).
         """
@@ -155,4 +153,4 @@ class RobertaTokenizer(GPT2Tokenizer):
 
         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]