Fixing question repetition ? - What wind.surf will do ?

app.py

@@ -164,15 +164,16 @@ with gr.Blocks() as demo:
     
     def submit_message(message, chat_history, model_name, system_message, max_tokens, temperature, top_p):
         history = [] if chat_history is None else chat_history
+        current_response = ""
         
-        # Add user message first
-        history = history + [{"role": "user", "content": message}]
-        
-        # Then stream the assistant's response
-        for response in respond(message, history[:-1], model_name, system_message, max_tokens, temperature, top_p):
-            history[-1] = {"role": "user", "content": message}
-            history = history + [{"role": "assistant", "content": response}]
-            yield history, ""
+        # Stream the assistant's response
+        for response in respond(message, history, model_name, system_message, max_tokens, temperature, top_p):
+            current_response = response
+            new_history = history + [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": current_response}
+            ]
+            yield new_history, ""
     
     submit_event = submit.click(
         fn=submit_message,