Update src/app/model.py

src/app/model.py

@@ -34,7 +34,7 @@ def load_model_and_processor(model_name: str, device: str) -> Any:
         # Load the model and processor
         model = (
             PaliGemmaForConditionalGeneration.from_pretrained(
-                model_name, token=access_token
+                model_name, torch_dtype=torch.bfloat16, token=access_token
             )
             .eval()
             .to(device)