Update app.py

app.py

@@ -6,11 +6,8 @@ import time
 
 import huggingface_hub as hf_hub
 # 初始化 OpenVINO 模型
-model_id = "hsuwill000/Llama-3.1-TAIDE-LX-8B-Chat_int4_ov"
+model_id = "OpenVINO/Qwen2-0.5B-int4-ov"
 model_path = "ov"
-#model_id = "OpenVINO/Phi-4-mini-instruct-int4-ov"
-#model_path = "Phi-4-mini-instruct-int4-ov"
-
 
 
 hf_hub.snapshot_download(model_id, local_dir=model_path)
@@ -30,7 +27,12 @@ def finish_chat():
 	
 # 建立推論函式：使用 streamer 並回傳 generator 結果
 def generate_stream(prompt):
-    prompt = prompt + " /no_think" + " 答案短且明瞭"
+    prompt = prompt #+ " /no_think" + " 答案短且明瞭"
+    chat = [
+        {"role": "user", "content": "{prompt}"},
+    ]
+    prompt = tokenizer.apply_chat_template(chat)
+    
     q = queue.Queue()
     tps_result = ""
     def streamer(subword):