Update app.py

app.py

@@ -23,7 +23,7 @@ def generate_stream(prompt):
 
     def worker():
         # 在背景 thread 中做推論
-        pipe.generate([prompt], streamer=streamer, max_new_tokens=512)
+        pipe.generate([prompt], streamer=streamer, max_new_tokens=32768)
         q.put(None)  # 結束符號
 
     threading.Thread(target=worker).start()
@@ -40,7 +40,7 @@ def generate_stream(prompt):
 with gr.Blocks() as demo:
     gr.Markdown("## 🧠 OpenVINO Streaming Demo with Gradio Textbox")
 
-    textbox_input = gr.Textbox(label="Prompt", lines=3, placeholder="Enter prompt here...")
+    textbox_input = gr.Textbox(label="Prompt", lines=1, placeholder="Enter prompt here...")
     textbox_output = gr.Textbox(label="Output", lines=10)
 
     # 按鈕控制觸發推論