Update handler.py

handler.py

@@ -3,7 +3,10 @@ from gliner import GLiNER
 class EndpointHandler:
     def __init__(self, path=""):
         # Use the provided path for loading the model
-        self.model = GLiNER.from_pretrained(path)
+        self.model = GLiNER.from_pretrained(
+            path,
+            device_map="cuda"
+        )
 
     def __call__(self, data):
     # If data is wrapped in 'inputs' (as Hugging Face does), unwrap it