dstack

dstack is an open-source orchestration layer that provisions GPUs and runs AI training, development environments and inference across GPU clouds, Kubernetes and on-prem clusters. ML teams use it as a simpler alternative to managing Kubernetes or Slurm by hand.

You describe dev environments, tasks, services and fleets in configuration files, and dstack finds capacity on the backends you connect, starts the jobs and tears them down afterwards. The same setup works whether the compute comes from a GPU cloud, a Kubernetes cluster or your own machines.

Because it targets heterogeneous compute, teams can mix providers and hardware vendors to find available or cheaper GPUs without rewriting their workflows. The core project is open source and can be self-hosted.