update for v5 - not exactly a GPT2Tokenizer

tokenizer_config.json

@@ -184,6 +184,6 @@
   "eos_token": "<|endoftext|>",
   "model_max_length": 8192,
   "pad_token": "<|pad|>",
-  "tokenizer_class": "GPT2Tokenizer",
+  "tokenizer_class": "PreTrainedTokenizerFast",
   "unk_token": "<|endoftext|>"
 }