Try to fix repeating on User

demo.py

@@ -98,18 +98,18 @@ def chat():
         if not message or (message == RETRY_COMMAND and len(chat_history) == 0):
             yield chat_history
             return
-
+    
         if message == RETRY_COMMAND and chat_history:
             prev_turn = chat_history.pop(-1)
             user_message, _ = prev_turn
             message = user_message
-
+    
         prompt = format_chat_prompt(message, chat_history, instructions)
         chat_history = chat_history + [[message, ""]]
         stream = llm(
             prompt,
             max_new_tokens=1024,
-            stop=[STOP_STR, "<|endoftext|>"],
+            stop=[STOP_STR, ""],
             temperature=temperature,
             top_p=top_p,
             stream=True
@@ -117,20 +117,27 @@ def chat():
         acc_text = ""
         for idx, response in enumerate(stream):
             text_token = response
-
-            if text_token in STOP_SUSPECT_LIST:
-                acc_text += text_token
+    
+            if acc_text.endswith(STOP_STR):
+                last_turn = list(chat_history.pop(-1))
+                last_turn[-1] += acc_text[:-len(STOP_STR)]
+                chat_history = chat_history + [last_turn]
+                yield chat_history
+                acc_text = text_token
                 continue
-
+    
             if idx == 0 and text_token.startswith(" "):
                 text_token = text_token[1:]
-
+    
             acc_text += text_token
+    
+        # If there's any remaining text after the loop
+        if acc_text:
             last_turn = list(chat_history.pop(-1))
             last_turn[-1] += acc_text
             chat_history = chat_history + [last_turn]
             yield chat_history
-            acc_text = ""
+
 
     def delete_last_turn(chat_history):
         if chat_history: