Amazon SageMaker Components for Kubeflow Pipelines
Today, we announced the public preview of Amazon SageMaker Components for Kubeflow Pipelines. Machine learning (ML) developers using Kubeflow Pipelines can convert their existing pipeline steps to run on SageMaker with the SageMaker Components. For instance, ML teams can use SageMaker for managed training on Spot instances which will automatically set up model checkpoints to S3 so that you can pause and resume training from the last saved state. Other SageMaker features that are supported in Kubeflow Pipelines are built-in algorithms, managed distributed training, and hyperparameter tuning. In addition, SageMaker can change instance types with one parameter swap, replacing the complicated autoscaling config in Kubernetes.
Amazon SageMaker is a fully managed service that is highly available, scalable, and reliable. SageMaker simplifies the infrastructure required to run a Kubeflow Pipeline environment. SageMaker Components for Kubeflow Pipelines currently support SageMaker Ground Truth, training, hyperparameter optimization, model creation, batch inference, and model endpoint creation jobs.
The Amazon SageMaker Components for Kubeflow Pipelines public preview is available in all regions where Amazon SageMaker is offered. Refer to AWS Region Table for details. To get started, visit the SageMaker Components for Kubeflow Pipelines documentation page.