Update tokenizer_config.json

This fixes the tokenizer to work with Transformers V5 otherwise you get the incorrect normalization.

See https://github.com/ml-explore/mlx-lm/issues/725 and https://github.com/huggingface/transformers/issues/43122 for relevant discussion.

tokenizer_config.json

@@ -490,6 +490,6 @@
   "clean_up_tokenization_spaces": false,
   "eos_token": "[e~[",
   "model_max_length": 40960000,
-  "tokenizer_class": "GPT2Tokenizer",
+  "tokenizer_class": "TokenizersBackend",
   "unk_token": "]!d~["
 }