Update handler.py

handler.py

@@ -4,9 +4,10 @@ import io
 
 class EndpointHandler:
     def __init__(self, model_dir=None):
-        # If a model_dir is provided, use it; otherwise, default to 'large-v2'.
+        # Set model size, assuming installation has been done with appropriate model files and setup
         model_size = "large-v2" if model_dir is None else model_dir
-        self.model = WhisperModel(model_size, device="cpu")
+        # Change to 'cuda' to use the GPU, and set compute_type for faster computation
+        self.model = WhisperModel(model_size, device="cuda", compute_type="float16")
 
     def __call__(self, data: Dict) -> Dict[str, str]:
         # Process the input data expected to be in 'inputs' key containing audio file bytes