Get scoring script from Azure ML Datastores

[mlopezfernandez]

• Package Name: azure.ai.ml
• Package Version: 1.21.1
• Operating System: Ubuntu 22.04
• Python Version: 3.10

Describe the bug According to the documentation, when you create a batch deployment it is possible to specify a scoring script from a local path, or "http:", "https:", or "azureml:". In my case when I specify it from a local path works correctly, but when I a specify a remote location does not work.

To Reproduce Steps to reproduce the behavior:

Instead of use a local path when you deploy a batch endpoint use a path from ML datastores or blob storage.

deployment = ModelBatchDeployment(
        name=model.split('.')[0].replace('_', '-') + "-1",
        endpoint_name=model.split('.')[0].replace('_', '-').lower(),
        model=Model(path="azureml://datastores/models/paths/" +
                    model, type=AssetTypes.CUSTOM_MODEL, name=model.split('.')[0]),
        code_configuration=CodeConfiguration(
            code="azureml://datastores/inference/paths/", scoring_script="inference.py"),
        environment=Environment(
            name="deploy-env",
            conda_file="../environment/inference-conda.yml",
            image="mcr.microsoft.com/azureml/openmpi4.1.0-ubuntu20.04:latest"
        ),
        compute="cpu-cluster",
        settings=ModelBatchDeploymentSettings(
            max_concurrency_per_instance=2,
            mini_batch_size=10,
            instance_count=2,
            output_action=BatchDeploymentOutputAction.APPEND_ROW,
            output_file_name="predictions.csv",
            retry_settings=BatchRetrySettings(max_retries=3, timeout=30),
            logging_level="info"
        )
    )

    ml_client.begin_create_or_update(deployment).result()

Screenshots

[kashifkhan]

Thank you for the feedback @mlopezfernandez . We will investigate and get back to you asap.

cc @azureml-github

[pracsol]

Any update on this? Still seems to be failing.