Update app.py

app.py

@@ -63,10 +63,10 @@ async def respond_thinking(message, history, selected_model):
 async def process_message(message, history, model_name):
     if "thinking" in model_name:
         async for response, thinking in respond_thinking(message, history, model_name):
-            yield response
+            yield (response, thinking)
     else:
         async for response in respond(message, history, model_name):
-            yield response
+            yield (response, "")
 
 def clear_thinking():
     return ""
@@ -96,7 +96,7 @@ with gr.Blocks() as demo:
         return gr.ChatInterface.update()
 
     async def update_thinking(history):
-        if history and history[-1][0] == "assistant" and len(history[-1]) > 1:
+        if history and history[-1][0]["role"] == "assistant" and len(history[-1]) > 1:
             return history[-1][1]
         return ""
 
@@ -106,7 +106,7 @@ with gr.Blocks() as demo:
         outputs=[chatbot],
     )
 
-    chatbot.on_message(
+    chatbot.change(
         update_thinking,
         inputs=[chatbot],
         outputs=[thinking_output]