Update tokenizer_class to LlamaTokenizerFast for GGUF compatibility

tokenizer_config.json

@@ -16,6 +16,6 @@
   ],
   "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<|endoftext|>",
-  "tokenizer_class": "TokenizersBackend",
+  "tokenizer_class": "LlamaTokenizerFast",
   "unk_token": "<|endoftext|>"
-}
+}