dstack – an alternative to Kubernetes and Slurm for AI

Abstract
ML teams need a simpler, more efficient way to manage AI infrastructure—provisioning GPUs and orchestrating development, training, and inference workloads. Kubernetes is often too complex and low-level, while Slurm lacks support for provisioning and development. dstack is an open-source project designed to fill these gaps and redefine how AI infrastructure is orchestrated.

About the Speaker
Andrey is the founder and core contributor to dstack. Before dstack, he spent over a decade at JetBrains building tools for developers.