Update app.py

app.py

@@ -26,10 +26,12 @@ def regenerate(_chatbot, _task_history, system_msg, max_tokens, temperature, top
     if not _task_history:
         yield _chatbot
         return
-    # Remove the last bot message
-    last_message = _task_history[-1]
-    if last_message[1]:  # Check if the bot actually sent a message last
-        last_message[1] = ""
+    # Remove the last bot message from history
+    last_message = _task_history.pop(-1)
+    _chatbot.pop(-1)
+    
+    # Re-add the user's part of the last message to history
+    _task_history.append([last_message[0], ""])
     
     # Regenerate the message using the chat function
     new_history, _, _ = chat(_task_history, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty)
@@ -39,6 +41,7 @@ def regenerate(_chatbot, _task_history, system_msg, max_tokens, temperature, top
 
 
 
+
 def chat(history, system_message, max_tokens, temperature, top_p, top_k, repetition_penalty):
     history = history or []
     
@@ -139,5 +142,6 @@ with gr.Blocks() as demo:
         fn=regenerate, inputs=[chatbot, chat_history_state, system_msg, max_tokens, temperature, top_p, top_k, repetition_penalty], outputs=[chatbot], queue=True
     )
 
+
 demo.queue(max_size=128, concurrency_count=2)
 demo.launch()