LocalStack Pro provides a local version of the SageMaker API, which allows running jobs to create machine learning models (e.g., using TensorFlow).

A basic example using the sagemaker.tensorflow.TensorFlow class is provided in this Github repository. Essentially, the code boils down to these core lines:

inputs = ...  # load training data files
mnist_estimator = TensorFlow(entry_point='', role='arn:aws:...',
    framework_version='1.12.0', sagemaker_session=sagemaker_session,
    train_instance_count=1, training_steps=10, evaluation_steps=10), logs=False)

The code snippet above uploads the model training code to local S3, submits a new training job to the local SageMaker API, and finally puts the trained model back to an output S3 bucket. Please refer to the sample repo for more details.

Last modified October 8, 2021: rename Local AWS Services to aws (fa6b2e4a)