This message was deleted.

what framework do you train your model with?

sklearn

then you should use

bentoml.sklearn

Currently, we don’t directly support azure ml, but you can always download the artifact locally, then use bentoml.sklearn to save it

Can you please elaborate on artifacts to be downloaded? Is it packages to be installed before loading the pickle file trained in azure ml environment?

Not sure how you setup your environment, Afaik, if you have access to the artifact that you trained you should able to save it with

bentoml.sklearn

Just make sure the sklearn and python version matches while serving using bentoml, I do it using a custom runner as such

def convert_model_to_runner(model: BaseEstimator):
    """Function to convert sklearn model to a bentoml.Runner instance to be used for serving.

    Args:
        model: Any sklearn model stored locally with serve code.
            Types allowed are BaseEstimator

    Returns:
        Bentoml.Runner instance which can be used to create a service.

    Raises:
        UnableToConvertModel: When conversion of model to Bentoml.Runner fails
    """
    try:
        prediction_runner = bentoml.Runner(
            ScikitLearnModelRunnable,
            name="prediction_runner",
            runnable_init_params={"model": model},
        )
        return cast(ScikitLearnModelRunner, prediction_runner)
    except TypeError as error:
        logger.error("Could not save model into a bentoml.Runner.", exc_info=True)
        raise UnableToConvertModel("Error mapping model to Bentoml.Runner") from error