Update tokenizer_config.json

(so it loads the right tokenizer class with our AutoTokenizer)

tokenizer_config.json

@@ -17,7 +17,7 @@
   "pad_token": "<pad>",
   "sep_token": "</s>",
   "special_tokens_map_file": null,
-  "tokenizer_class": "RobertaTokenizer",
+  "tokenizer_class": "GPT2Tokenizer",
   "trim_offsets": true,
   "unk_token": "<unk>"
 }