Update app.py

app.py

@@ -6,9 +6,9 @@ import numpy as np
 import gradio as gr
 import re
 import gc
+from openvino_genai import GenerationConfig
 
-
-hf_hub.snapshot_download(repo_id="OpenVINO/Qwen3-0.6B-int4-ov", local_dir="ov", local_dir_use_symlinks=False)
+hf_hub.snapshot_download(repo_id="hsuwill000/Llama-3.1-TAIDE-LX-8B-Chat_int4_ov", local_dir="ov", local_dir_use_symlinks=False)
 
 # 初始化模型
 device = "CPU"
@@ -18,7 +18,10 @@ pipe = ov_genai.LLMPipeline(model_path, device)
 
 tokenizer = pipe.get_tokenizer()
 tokenizer.set_chat_template(tokenizer.chat_template)
-
+# 定義你要偵測的結束詞（注意是最終解碼後出現的文字）
+config = GenerationConfig(
+    stop_strings=set(["<|eot_id|>"])  # ✅ 這是 set
+)
 
 def streamer(subword):
     global output_buffer