Update tokenizer_config.json

This should make it compatible with v5 + corrects it because it does not follow gpt2.
GPT2 does not have a normalizer, and has a different pre-tokenizer https://huggingface.co/openai-community/gpt2/raw/main/tokenizer.json

tokenizer_config.json

@@ -491,7 +491,7 @@
   "eos_token": "[e~[",
   "extra_special_tokens": {},
   "model_max_length": 40960000,
-  "tokenizer_class": "GPT2Tokenizer",
+  "tokenizer_class": "TokenizersBackend",
   "unk_token": "]!d~[",
   "tool_parser_type": "minimax_m2"
 }