diff --git a/src/llmtuner/chat/stream_chat.py b/src/llmtuner/chat/stream_chat.py
index b1ded67a..f39ed960 100644
--- a/src/llmtuner/chat/stream_chat.py
+++ b/src/llmtuner/chat/stream_chat.py
@@ -1,7 +1,8 @@
 import torch
+from types import MethodType
 from typing import Any, Dict, Generator, List, Optional, Tuple
 from threading import Thread
-from transformers import TextIteratorStreamer
+from transformers import PreTrainedModel, TextIteratorStreamer
 
 from llmtuner.extras.misc import dispatch_model, get_logits_processor, get_stopwords_criteria
 from llmtuner.extras.template import get_template
@@ -20,6 +21,7 @@ class ChatModel:
             self.tokenizer.encode(word, add_special_tokens=False)[0] for word in self.template.stop_words
         ]
         self.tokenizer.add_special_tokens(dict(additional_special_tokens=self.template.stop_words))
+        self.model.generate = MethodType(PreTrainedModel.generate, self.model) # a monkey fix for qwen model
 
     def process_args(
         self,