diff --git a/tools/fastllm_pytools/hf_model.py b/tools/fastllm_pytools/hf_model.py
index 439227e3..f2135ff7 100644
--- a/tools/fastllm_pytools/hf_model.py
+++ b/tools/fastllm_pytools/hf_model.py
@@ -64,6 +64,10 @@ def create(model,
             modelInfo["im_start_id"] = tokenizer.im_start_id
     elif (modelInfo["model_type"] == "qwen2"):
         modelInfo["eos_token_id"] = "151645"
+        pre_prompt = modelInfo["pre_prompt"] = "<|im_start|>system\nYou are a helpful assistant.<|im_end|>"
+        user_role = modelInfo["user_role"] = "<|im_start|>user\n"
+        bot_role = modelInfo["bot_role"] = "<|im_end|><|im_start|>assistant\n"
+        history_sep = modelInfo["history_sep"] = "<|im_end|>\n"
     elif (modelInfo["model_type"] == "internlm"):
         modelInfo["eos_token_id"] = "103028"
         if "rotary" in modelInfo:
@@ -81,7 +85,6 @@ def create(model,
         rope_scaling = modelInfo.pop("rope_scaling")
         modelInfo["rope_scaling.type"] = rope_scaling["type"]
         modelInfo["rope_scaling.factor"] = rope_scaling["factor"]
-
     if tokenizer:
         modelInfo["tokenizer_use_score"] = "1" # 分词带分数
         if len(tokenizer.all_special_tokens) > 0:
diff --git a/tools/fastllm_pytools/llm.py b/tools/fastllm_pytools/llm.py
index 75024a4c..29d079bb 100644
--- a/tools/fastllm_pytools/llm.py
+++ b/tools/fastllm_pytools/llm.py
@@ -351,7 +351,9 @@ def stream_chat(self, tokenizer, query: str, history: List[Tuple[str, str]] = No
                     max_length: int = 8192, do_sample = True, top_p = 0.8, top_k = 1, temperature = 1.0, repeat_penalty = 1.0,
                     return_past_key_values = False, stop_token_ids: List[int] = None, **kwargs) -> str:
         type = None
-        if (tokenizer.name == "GLMTokenizer" and hasattr(tokenizer, "build_chat_input")):
+        if (hasattr(tokenizer, "name") 
+            and tokenizer.name == "GLMTokenizer" 
+            and hasattr(tokenizer, "build_chat_input")):
             type = "ChatGLM3"
 
         if (not(history)):