Update app.py

app.py

@@ -20,6 +20,8 @@ def start_chat():
 def finish_chat():
     pipe.finish_chat()
     return "🛑 結束對話！"
+
+
 	
 # 建立推論函式：使用 streamer 並回傳 generator 結果
 def generate_stream(prompt):
@@ -32,7 +34,13 @@ def generate_stream(prompt):
 
     def worker():
         # 在背景 thread 中做推論
-        pipe.generate([prompt], streamer=streamer, max_new_tokens=4096)
+		try
+            gen_result = pipe.generate([prompt], streamer=streamer, max_new_tokens=4096)
+            tps = gen_result.perf_metrics.get_throughput().mean
+            result_container["tps"] = f"{tps:.2f} tokens/s"
+        except Exception as e:
+            result_container["tps"] = f"❌ {str(e)}"
+			
         q.put(None)  # 結束符號
 
     threading.Thread(target=worker).start()
@@ -55,11 +63,13 @@ with gr.Blocks(css="""
 }
 """) as demo:
     gr.Markdown("## 🧠 OpenVINO Streaming Demo with Gradio Textbox")
-	
-    start_btn = gr.Button("開始對話")
-    end_btn = gr.Button("結束對話")
-
-    status_box = gr.Textbox(label="狀態", interactive=False)
+    
+    with gr.Row():
+        start_btn = gr.Button("開始對話")
+        end_btn = gr.Button("結束對話")
+        status_box = gr.Textbox(label="狀態", interactive=False)
+        TPS_box = gr.Textbox(label="TPS", interactive=False)
+    
     start_btn.click(fn=start_chat, outputs=status_box)
     end_btn.click(fn=finish_chat, outputs=status_box)
 	
@@ -72,4 +82,4 @@ with gr.Blocks(css="""
     # 當按鈕被按下時，呼叫 generate_stream 並更新 textbox_output
     button.click(fn=generate_stream, inputs=textbox_input, outputs=textbox_output)
 
-demo.launch()
+demo.launch()