fix: fix bug with cuda devices in clip-nebullvm

server/clip_server/model/clip_nebullvm.py

@@ -77,17 +77,21 @@ class EnvRunner:
     def __init__(self, device: str, num_threads: int = None):
         self.device = device
         self.cuda_str = None
+        self.rm_cuda_flag = False
         self.num_threads = num_threads
 
     def __enter__(self):
         if self.device == "cpu" and torch.cuda.is_available():
-            self.cuda_str = os.environ.get("CUDA_VISIBLE_DEVICES") or "1"
+            self.cuda_str = os.environ.get("CUDA_VISIBLE_DEVICES")
             os.environ["CUDA_VISIBLE_DEVICES"] = "0"
+            self.rm_cuda_flag = self.cuda_str is None
         if self.num_threads is not None:
             os.environ["NEBULLVM_THREADS_PER_MODEL"] = f"{self.num_threads}"
 
     def __exit__(self, exc_type, exc_val, exc_tb):
         if self.cuda_str is not None:
             os.environ["CUDA_VISIBLE_DEVICES"] = self.cuda_str
+        elif self.rm_cuda_flag:
+            os.environ.pop("CUDA_VISIBLE_DEVICES")
         if self.num_threads is not None:
             os.environ.pop("NEBULLVM_THREADS_PER_MODEL")