Update app.py

app.py

@@ -1,3 +1,4 @@
+
 import time
 import gradio as gr
 from huggingface_hub import InferenceClient
@@ -12,53 +13,41 @@ tokenizer = AutoTokenizer.from_pretrained(model_id)
 # 建立生成管道
 pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
 
-# 回應生成器
-def respond_stream(message):
+def respond(message, history):
     start_time = time.time()  # 記錄開始時間
 
+    # 將當前訊息與歷史訊息合併
+    input_text = message if not history else history[-1]["content"] + " " + message
+    input_text = message
     # 設定生成參數
-    max_length = 1024
-    stop_token = "<|endoftext|>"
-    generated_text = ""
-
-    # 使用生成器逐步輸出
-    for response in pipe(message, max_length=max_length, truncation=False, num_return_sequences=1, stream=True):
-        new_text = response["generated_text"]
-        generated_text += new_text
-        yield generated_text  # 將部分生成結果發送給 Gradio 界面
-
-        # 偵測結束條件
-        if stop_token in new_text:
+    max_length = 1024  # 增加最大生成長度
+    output_text = ""
+    stop_token = "<|endoftext|>"  # 可選：結束標記
+    while True:
+        response = pipe(input_text, max_length=max_length, truncation=False, num_return_sequences=1)
+        reply = response[0]['generated_text']
+        output_text += reply
+        
+        # 檢測是否包含結束標記，或者生成結束
+        if stop_token in reply or len(output_text) >= max_length:
+            output_text = output_text.split(stop_token)[0]  # 去掉結束標記以後的部分
             break
+        
+        # 更新輸入文字繼續生成
+        input_text = reply
 
-    end_time = time.time()
-    print(f"Time taken: {end_time - start_time:.2f} seconds")
-
-# 聊天接口函數
-def chatbot_interface(message, history):
-    history = history or []
-    history.append({"role": "user", "content": message})
-    response = respond_stream(message)
-    for partial_response in response:
-        history.append({"role": "assistant", "content": partial_response})
-        yield history  # 實時更新歷史紀錄
+    end_time = time.time()  # 記錄結束時間
+    duration = end_time - start_time  # 計算耗時
 
-# 清除聊天紀錄
-def clear_chat():
-    return [], ""
+    # 輸出耗時到控制台
+    print(f"Message: {message}")
+    print(f"Reply: {output_text}")
+    print(f"Time taken for response: {duration:.2f} seconds")
 
-# Gradio 界面設計
-with gr.Blocks() as demo:
-    chatbot = gr.Chatbot(label="Chat with Model")
-    msg = gr.Textbox(label="Enter your message", placeholder="Type your message here...", lines=2)
-    send = gr.Button("Send")  # 顯示送出按鈕
-    clear = gr.Button("Clear")  # 清除按鈕
+    return output_text
 
-    # 配置交互
-    send.click(chatbot_interface, inputs=[msg, chatbot], outputs=chatbot)
-    msg.submit(chatbot_interface, inputs=[msg, chatbot], outputs=chatbot)  # 支援按下Enter
-    clear.click(clear_chat, inputs=[], outputs=[chatbot, msg])  # 清除按鈕功能
+# 設定 Gradio 的聊天界面
+demo = gr.ChatInterface(fn=respond, title="Phi-3.5-mini-instruct-openvino", description="Phi-3.5-mini-instruct-openvino", type='messages')
 
-# 啟動
 if __name__ == "__main__":
     demo.launch()