Update app.py

app.py

@@ -18,25 +18,39 @@ tokenizer = pipe.get_tokenizer()
 tokenizer.set_chat_template(tokenizer.chat_template)
 
 
+def streamer(subword):
+    yield subword
+    return ov_genai.StreamingStatus.RUNNING
+
+
 def generate_response(prompt):
     try:
-        generated = pipe.generate([prompt], max_length=1024)
-        tokenpersec=f'{generated.perf_metrics.get_throughput().mean:.2f}'
+        full_response = ""
+        token_count = 0
+        start_time = time.time()
+
+        for text in pipe.generate(prompt, streamer=streamer, max_new_tokens=1024):
+            full_response += text
+            token_count += 1
+            yield (None, full_response) # 每次 yield 都会刷新界面
+
+        end_time = time.time()
+        elapsed_time = end_time - start_time
+        tokens_per_sec = token_count / elapsed_time if elapsed_time > 0 else 0
+        tokenpersec=f'{tokens_per_sec:.2f}'
+
+        yield (tokenpersec, full_response)  # 最终 yield, 保证输出完整.
 
-        return tokenpersec, generated
     except Exception as e:
-        return "發生錯誤", "發生錯誤", f"生成回應時發生錯誤：{e}"
-    
+        yield ("發生錯誤", f"生成回應時發生錯誤：{e}")  # 使用 yield 错误信息
 
 # 建立 Gradio 介面
 demo = gr.Interface(
     fn=generate_response,
     inputs=gr.Textbox(lines=5, label="輸入提示 (Prompt)"),
     outputs=[
-        gr.Textbox(label="tokens/sec"),        
-        #gr.Textbox(label="思考過程"),
-        #gr.Textbox(label="最終回應")
-        gr.Textbox(label="回應")
+        gr.Textbox(label="tokens/sec"),
+        gr.Textbox(label="回應", streaming=True)
     ],
     title="Qwen3-0.6B-int4-ov ",
     description="基於 Qwen3-0.6B-int4-ov 推理應用，支援思考過程分離與 GUI。"