Ray on SageMaker HyperPod
Ray is an open source unified framework for scaling AI and Python applications. It
provides the compute layer for parallel processing, so you do not need to be a distributed
systems expert to run work across a cluster. Ray Core distributes the work, and the Ray AI
libraries cover data processing, distributed training, hyperparameter tuning, reinforcement
learning, and model serving. For more information, see Overview
Ray on SageMaker HyperPod runs your Ray workloads on HyperPod compute orchestrated by Amazon EKS, using the open source KubeRay operator. Your Ray code, the Ray APIs, and the KubeRay custom resources are unchanged.
Staying fully aligned with open source Ray, HyperPod adds capabilities around it rather than replacing any part of it:
-
A purpose-built web interface to operate Ray workloads in Amazon SageMaker Studio that brings all the capabilities below under a single interface.
-
Interactive development, where a JupyterLab, Code Editor, or remote IDE session attaches to a Ray cluster and
ray.init()connects. -
Secure Ray Dashboard access through an authenticated browser link, without
kubectl port-forward. -
A managed tiered KV cache that lowers time to first token, JumpStart model deployment, and managed Karpenter autoscaling for Ray Serve.
-
Team compute allocations, priority-based scheduling, and lending and borrowing of idle compute through HyperPod Task Governance.
-
Ray metrics and Grafana dashboards provisioned by the HyperPod Observability add-on.
-
Resilient training with automatic node recovery, hung job detection, and tiered checkpointing at the infrastructure layer.
You can adopt all of this, or install only the capabilities you are missing. For more information about the two approaches, see Getting started.
Prerequisites
Ray on HyperPod requires a HyperPod cluster orchestrated by Amazon EKS. For more information about creating one, see Creating a SageMaker HyperPod cluster with Amazon EKS orchestration.
Regions
Ray on HyperPod is available in all AWS Regions where SageMaker HyperPod supports Amazon EKS orchestration. For the list of Regions, see AWS Regions supported by SageMaker HyperPod.