Update app.py

app.py

@@ -18,43 +18,49 @@ 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, history=[]):  # Added history parameter
+    full_prompt = tokenizer.apply_chat_template(
+        messages=history + [{"role": "user", "content": prompt}],
+        tokenize=False,
+        add_generation_prompt=True
+    )
 
-
-def generate_response(prompt):
-    try:
+    # Create a generator function for streaming
+    def stream_generate():
         full_response = ""
-        token_count = 0
         start_time = time.time()
+        token_count = 0
 
-        for text in pipe.generate(prompt, streamer=streamer, max_new_tokens=1024):
-            full_response += text
+        def streamer(subword):
+            nonlocal full_response, token_count
+            full_response += subword
             token_count += 1
-            yield (None, full_response) # 每次 yield 都会刷新界面
+            yield full_response, f'{token_count / (time.time() - start_time):.2f}'
+            return ov_genai.StreamingStatus.RUNNING
+        
+        try:
+            pipe.generate(full_prompt, streamer=streamer, max_new_tokens=1024)
+            # Add user prompt and full response to the history
+            history.append({"role": "user", "content": prompt})
+            history.append({"role": "assistant", "content": full_response})
+        except Exception as e:
+            yield "發生錯誤", f"生成回應時發生錯誤：{e}"
+        
+        # Yield the final response and tokens/sec
+        yield full_response, f'{token_count / (time.time() - start_time):.2f}'
 
-        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.
-
-    except Exception as e:
-        yield ("發生錯誤", f"生成回應時發生錯誤：{e}")  # 使用 yield 错误信息
+    return stream_generate()
 
 # 建立 Gradio 介面
-demo = gr.Interface(
+demo = gr.ChatInterface(
     fn=generate_response,
-    inputs=gr.Textbox(lines=5, label="輸入提示 (Prompt)"),
-    outputs=[
-        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。"
+    description="基於 Qwen3-0.6B-int4-ov 推理應用，支援思考過程分離與 GUI.",
+    additional_inputs=[
+        gr.Textbox(label="tokens/sec", value="", interactive=False)
+    ]
 )
 
+
 if __name__ == "__main__":
-    demo.launch()
+    demo.queue().launch()