try

app.py

@@ -30,6 +30,10 @@ MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
 ##
 # 检查CUDA是否可用
 def print_gpu():
+    result = subprocess.run(
+        ['nvidia-smi'], text=True
+    )
+
     if torch.cuda.is_available():
         print("CUDA is available. Listing available GPUs:")
         # 获取并打印GPU数量