Update handler.py

handler.py

@@ -2,11 +2,11 @@ from gliner import GLiNER
 
 class EndpointHandler:
     def __init__(self, path=""):
-        # Use the provided path for loading the model
-        self.model = GLiNER.from_pretrained(
-            path,
-            device_map="cuda"
-        )
+        # Load without device_map, then move to GPU
+        self.model = GLiNER.from_pretrained(path)  # Remove device_map="cuda"
+        device = "cuda" if torch.cuda.is_available() else "cpu"
+        self.model = self.model.to(device)
+        self.model.eval()  # Lock for inference
 
     def __call__(self, data):
     # If data is wrapped in 'inputs' (as Hugging Face does), unwrap it