fix requires for windows

src/llmtuner/chat/hf_engine.py

@@ -35,7 +35,7 @@ class HuggingfaceEngine(BaseEngine):
         self.template = get_template_and_fix_tokenizer(self.tokenizer, data_args.template)
         self.model = load_model(
             self.tokenizer, model_args, finetuning_args, is_trainable=False, add_valuehead=(not self.can_generate)
-        )
+        )  # must after fixing tokenizer to resize vocab
         self.generating_args = generating_args.to_dict()
 
     @staticmethod