Overview
Learn about a novel Hydra architecture approach for executing machine learning workflows across multiple Kubernetes clusters, regions, and clouds in this 24-minute OSACon conference talk. Discover how to overcome challenges like GPU availability constraints, multi-provider cloud credits management, and data residency requirements without resource-intensive replication methods. Explore an open source solution that enables standard Python programs to be dispatched to any compute resource while maintaining centralized management and orchestration in a single cluster. Master the technique of unbundling orchestration and execution to provide practitioners with a unified abstract layer for accessing distributed compute resources, eliminating the need for multiple deployment packages or environment-specific restrictions.
Syllabus
Hydra Architecture: Orchestrating ML across clusters, regions, and clouds
Taught by
OSACon