Introducing Amazon EKS support in Amazon SageMaker HyperPod

This article introduces the Amazon EKS support in Amazon SageMaker HyperPod, a purpose-built infrastructure designed for resilience and scalability in training foundation models (FMs).

Specifically, the article covers:

• Overview of Amazon EKS support in SageMaker HyperPod, including the architecture, managed resiliency features (deep health checks, automated node recovery, job auto resume), and user experiences for admins and scientists
• Detailed guide on setting up HyperPod compute as worker nodes in an EKS cluster, with emphasis on the built-in resiliency features like deep health checks and automated node recovery
• Demo of training job resiliency using the job auto resume functionality with the Kubeflow Training Operator for PyTorch, enabling automatic job recovery and continuation after interruptions or failures
• Conclusion highlighting the benefits of EKS support in SageMaker HyperPod for resilient and scalable FM training, and resources for further learning