Remove TokenizersBackend tokenizer_class for vLLM compatibility

tokenizer_config.json

@@ -180,6 +180,5 @@
     "attention_mask"
   ],
   "model_max_length": 1000000000000000019884624838656,
-  "pad_token": "<|endoftext|>",
-  "tokenizer_class": "PreTrainedTokenizerFast"
-}
+  "pad_token": "<|endoftext|>"
+}