Update app.py

app.py

@@ -9,60 +9,51 @@ import gc
 
 # 下載模型
 
+# 修正 download 路徑使用
 model_ids = [
     "OpenVINO/Qwen3-0.6B-int4-ov",
     "OpenVINO/Qwen3-1.7B-int4-ov",
-    #"OpenVINO/Qwen3-4B-int4-ov",#不可用
     "OpenVINO/Qwen3-8B-int4-ov",
     "OpenVINO/Qwen3-14B-int4-ov",
 ]
 
-model_name_to_full_id = {model_id.split("/")[-1]: model_id for model_id in model_ids}  #Create Dictionary
+model_name_to_full_id = {model_id.split("/")[-1]: model_id for model_id in model_ids}
 
 for model_id in model_ids:
-    model_path = model_id.split("/")[-1]  # Extract model name
+    model_path = model_id.split("/")[-1]
     try:
-      hf_hub.snapshot_download(model_id, local_dir=model_path, local_dir_use_symlinks=False)
-      print(f"Successfully downloaded {model_id} to {model_path}") # Optional: Print confirmation
+        hf_hub.snapshot_download(repo_id=model_id, local_dir=model_path, local_dir_use_symlinks=False)
+        print(f"Downloaded {model_id} to {model_path}")
     except Exception as e:
-      print(f"Error downloading {model_id}: {e}") # Handle download errors gracefully
+        print(f"Download error: {e}")
 
-
-#hf_hub.snapshot_download("hsuwill000/Llama-3.1-TAIDE-LX-8B-Chat_int4_ov", local_dir="ov", local_dir_use_symlinks=False)
-
-# 建立推理管線 (Initialize with a default model first)
+# 初始化模型
 device = "CPU"
-InUsed_model_name = "Qwen3-0.6B-int4-ov"  # Choose a default model
-pipe = ov_genai.LLMPipeline(InUsed_model_name, device)
+InUsed_model_name = "Qwen3-0.6B-int4-ov"
+model_path = f"./{InUsed_model_name}"  # 加上目錄路徑
+pipe = ov_genai.LLMPipeline(model_path, device)
 tokenizer = pipe.get_tokenizer()
-tokenizer.set_chat_template(tokenizer.chat_template)
-config = ov_genai.GenerationConfig()
-config.max_new_tokens = 4096
+config = ov_genai.GenerationConfig(max_new_tokens=4096)
 
-def streamer(subword):
-    print(subword, end='', flush=True)
-    return False
-    
+# 推理函式
 def generate_response(prompt, model_name):
     global pipe, tokenizer, InUsed_model_name
-    if(InUsed_model_name!=model_name):
-        model_path = model_name
+    if InUsed_model_name != model_name:
         del pipe
         gc.collect()
-        print(f"Switching to model: {model_name}")
+        model_path = f"./{model_name}"  # 指定正確本機資料夾
         pipe = ov_genai.LLMPipeline(model_path, device)
         tokenizer = pipe.get_tokenizer()
-        tokenizer.set_chat_template(tokenizer.chat_template)
-        InUsed_model_name=model_name
+        InUsed_model_name = model_name
 
     try:
         generated = pipe.generate([prompt], config, streamer)
-        tokenpersec=f'{generated.perf_metrics.get_throughput().mean:.2f}'
+        tokenpersec = f'{generated.perf_metrics.get_throughput().mean:.2f}'
         print(f"\nModel:{model_name} TPS:{tokenpersec}\n")
-        return tokenpersec, generated
+        return tokenpersec, generated.text
     except Exception as e:
-        return "發生錯誤", "發生錯誤", f"生成回應時發生錯誤：{e}"
-    
+        return "錯誤", f"生成失敗：{e}"
+
 
 # 建立 Gradio 介面
 model_choices = list(model_name_to_full_id.keys())