Update app.py

app.py

@@ -35,10 +35,13 @@ tokenizer.set_chat_template(tokenizer.chat_template)
 config = ov_genai.GenerationConfig()
 config.max_new_tokens = 1024
 
-def streamer(subword):
-    print(subword, end='', flush=True)
-    return False
+# 修改 streamer 函數，使其 yield 結果
+def streamer():
+    def cb(subword):
+        yield subword
+    return cb
     
+# 修改 generate_response 函數，使用 yield 傳回結果，並傳回tokenpersec
 def generate_response(prompt, model_name):
     global pipe, tokenizer, InUsed_model_name
     if(InUsed_model_name!=model_name):
@@ -52,12 +55,17 @@ def generate_response(prompt, model_name):
         InUsed_model_name=model_name
 
     try:
-        generated = pipe.generate([prompt], config, streamer)
-        tokenpersec=f'{generated.perf_metrics.get_throughput().mean:.2f}'
+        response = ""
+        callback = streamer()
+        for token in pipe.generate([prompt], config, callback):
+            response += token
+            yield '', response
+
+        tokenpersec = f'{pipe.last_metrics.get_throughput().mean:.2f}'
+        yield tokenpersec, response  # 傳回最終結果
 
-        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}"
     
 
 # 建立 Gradio 介面
@@ -71,11 +79,12 @@ demo = gr.Interface(
     ],
     outputs=[
         gr.Textbox(label="tokens/sec"),
-        gr.Textbox(label="回應", streaming=True),
+        gr.Textbox(label="回應"),
     ],
     title="Qwen3 Model Inference",
-    description="基於 Qwen3 推理應用，支援思考過程分離與 GUI。"
+    description="基於 Qwen3 推理應用，支援思考過程分離與 GUI。",
+    live=True  # 啟用即時更新
 )
 
 if __name__ == "__main__":
-    demo.launch()
+    demo.queue().launch()