[update]

main.py

@@ -137,12 +137,6 @@ def chat_with_llm_streaming(question: str,
 
     model, tokenizer = init_model(pretrained_model_name_or_path)
 
-    # input_ids
-    if model.config.model_type == "chatglm":
-        input_ids = []
-    else:
-        input_ids = [tokenizer.bos_token_id]
-
     # history
     utterances = list()
     for idx, (h_question, h_answer) in enumerate(history):
@@ -157,6 +151,12 @@ def chat_with_llm_streaming(question: str,
     encoded_utterances = tokenizer.__call__(utterances, add_special_tokens=False)
     encoded_utterances = encoded_utterances["input_ids"]
 
+    # input_ids
+    if model.config.model_type == "chatglm":
+        input_ids = []
+    else:
+        input_ids = [tokenizer.bos_token_id]
+
     for encoded_utterance in encoded_utterances:
         input_ids.extend(encoded_utterance)
         if model.config.model_type != "chatglm":