Update tokenizer_class to PreTrainedTokenizerFast for GGUF compatibility

tokenizer_config.json

@@ -18,6 +18,6 @@
   "local_files_only": false,
   "model_max_length": 1000000000000000019884624838656,
   "pad_token": "<|endoftext|>",
-  "tokenizer_class": "LlamaTokenizerFast",
+  "tokenizer_class": "PreTrainedTokenizerFast",
   "unk_token": "<|endoftext|>"
 }