[camembert] tokenizer: use additional_special_tokens

transformers/tokenization_camembert.py

@@ -45,10 +45,12 @@ class CamembertTokenizer(PreTrainedTokenizer):
     max_model_input_sizes = PRETRAINED_POSITIONAL_EMBEDDINGS_SIZES
 
     def __init__(self, vocab_file, bos_token="<s>", eos_token="</s>", sep_token="</s>",
-                 cls_token="<s>", unk_token="<unk>", pad_token='<pad>', mask_token='<mask>', **kwargs):
+                 cls_token="<s>", unk_token="<unk>", pad_token='<pad>', mask_token='<mask>',
+                 additional_special_tokens=['<s>NOTUSED', '<s>NOTUSED'], **kwargs):
         super(CamembertTokenizer, self).__init__(max_len=512, bos_token=bos_token, eos_token=eos_token, unk_token=unk_token,
                                                  sep_token=sep_token, cls_token=cls_token, pad_token=pad_token,
-                                                 mask_token=mask_token, **kwargs)
+                                                 mask_token=mask_token, additional_special_tokens=additional_special_tokens,
+                                                 **kwargs)
         self.max_len_single_sentence = self.max_len - 2  # take into account special tokens
         self.max_len_sentences_pair = self.max_len - 4  # take into account special tokens
         self.sp_model = spm.SentencePieceProcessor()