TransformStep fails in local mode when using ModelPackage

[alex-vinnik-sp]

I am trying to execute pipeline with TransformStep in local mode

    model = ModelPackage(
        role=role,
        model_data=model_data,
        model_package_arn=model_package_arn,
    )
    model.sagemaker_session = sagemaker_session
    transformer = model.transformer(
        instance_count=1,
        instance_type=instance_type,
        strategy="MultiRecord",
        assemble_with="Line",
        accept="text/csv",
        output_path=output_path,
        env=env,
    )

    return TransformStep(
        name="ScoringData",
        description="Take LSO feature data and output a csv with outlier classifications using the latest model",
        transformer=transformer,
        inputs=TransformInput(
            data=data_input,
            content_type="text/csv",
            join_source="None",
            split_type="Line",
        ),
        depends_on=depends_on,
        retry_policies=RETRY_POLICIES,
    )

and it fails

  File "/Users/alex.vinnik/code/saas-mlops-ml-platform/pipelines/saas-ida-outliers-model-build/steps.py", line 355, in get_step_transform
    transformer = model.transformer(
  File "/Users/alex.vinnik/code/saas-mlops-ml-platform/.venv/lib/python3.10/site-packages/sagemaker/model.py", line 1504, in transformer
    self._create_sagemaker_model(instance_type, tags=tags)
  File "/Users/alex.vinnik/code/saas-mlops-ml-platform/.venv/lib/python3.10/site-packages/sagemaker/model.py", line 1864, in _create_sagemaker_model
    enable_network_isolation=self.enable_network_isolation(),
  File "/Users/alex.vinnik/code/saas-mlops-ml-platform/.venv/lib/python3.10/site-packages/sagemaker/model.py", line 1804, in enable_network_isolation
    return self._is_marketplace()
  File "/Users/alex.vinnik/code/saas-mlops-ml-platform/.venv/lib/python3.10/site-packages/sagemaker/model.py", line 1816, in _is_marketplace
    model_package_desc = sagemaker_session.sagemaker_client.describe_model_package(
AttributeError: 'LocalSagemakerClient' object has no attribute 'describe_model_package'

Looks like describe_model_package is not supported by LocalSagemakerClient. SDK version 2.173.
Is there a way to workaround it? Thanks.

[alex-vinnik-sp]

Looks like the following code gives you an agnostic pipeline suitable for both local and cloud

    model = Model(
        image_uri="image uri",
        model_data="uri to model.tar.gz",
        role=role,
        sagemaker_session=sagemaker_session,
    )

    step_create_model = ModelStep(name="Your name", step_args=model.create())

    transformer = Transformer(
        model_name=step_create_model.properties.ModelName,
        instance_type=instance_type,
        instance_count=1,
        sagemaker_session=sagemaker_session,
        env=env,
        output_path=output_path,
        strategy="MultiRecord",
        assemble_with="Line",
        accept="text/csv",
    )

    transform_args = transformer.transform(data_input, content_type="text/csv")

    TransformStep(
        name="Your name",
        description="whatever",
        step_args=transform_args,
        inputs=TransformInput(
            data=data_input,
        ),
        depends_on=depends_on,
    )