service.py

...
ONNX_MODEL = "onnx_geolocator"
VERSION = "latest"


class GeoLocatorRunnable(bentoml.Runnable):
    """
    Custom BentoML runner to fetch multiple ONNX outputs
    """

    SUPPORTED_RESOURCES = ("nvidia.com/gpu", "cpu")
    SUPPORTS_CPU_MULTI_THREADING = True

    def __init__(self):
        super().__init__()

        # load the model instance
        self.model = bentoml.onnx.load_model(f"{ONNX_MODEL}:{VERSION}")

    @bentoml.Runnable.method(batchable=False)
    def fetch_multiple_onnx_outputs(self, images) -> List[np.ndarray]:
        ort_inputs = {self.model.get_inputs()[0].name: to_numpy(images)}
        ort_outs = self.model.run(None, ort_inputs)
        return ort_outs


# initialize custom bentoml runner, svc and input spec
geolocator_runner = bentoml.Runner(
    GeoLocatorRunnable,
    models=[bentoml.onnx.get(f"{ONNX_MODEL}:{VERSION}")],
)
svc = bentoml.Service("geolocator", runners=[geolocator_runner])

...

@svc.api(input=Image(), output=Text(), route="predict-image")
def predict_image(image: PILImage) -> str:
    image_dir = img_processor(img_data=image)
    return predict_helper(image_dir=image_dir)