fix: change tokenizer_class to Qwen2TokenizerFast for vLLM compatibility

tokenizer_config.json

@@ -9,6 +9,6 @@
   "model_specific_special_tokens": {},
   "pad_token": "<|endoftext|>",
   "split_special_tokens": false,
-  "tokenizer_class": "Qwen2Tokenizer",
+  "tokenizer_class": "Qwen2TokenizerFast",
   "unk_token": null
 }