Update app.py

app.py

@@ -220,9 +220,9 @@ def clear_chat():
 def chat_fn(message, history, system_prompt, temperature):
     """Main chat function"""
     if not chat_model.model_loaded:
-        return "Please wait for the model to load or reload the space."
+        return history + [[message, "Please wait for the model to load or reload the space."]]
     
-    # Convert history format
+    # Convert history format for the model
     formatted_history = []
     for user_msg, assistant_msg in history:
         formatted_history.append({"role": "user", "content": user_msg})
@@ -232,7 +232,8 @@ def chat_fn(message, history, system_prompt, temperature):
     # Generate response
     response = chat_model.generate_response(message, formatted_history, system_prompt, temperature)
     
-    return response
+    # Return updated history with new message pair
+    return history + [[message, response]]
 
 def chat_stream_fn(message, history, system_prompt, temperature):
     """Streaming chat function"""