dstack

dstack is an open-source engine that automates infrastructure provisioning for developing, training and deploying AI models on a variety of cloud computing services and data centers. It lets you orchestrate AI workloads, whether that's running jobs, deploying models or setting up dev environments.

dstack is geared for complex AI workflows, with features including:

• Dev environments: Spin up remote machines with your favorite IDE and resources for interactive coding.
• Tasks: Schedule batch jobs for training, fine-tuning or data processing on individual machines or clusters.
• Services: Deploy models as scalable and secure endpoints for easy public access.
• Pools: Repurpose cloud instances and on-premises servers for efficient reuse and easier management.

dstack supports a variety of cloud computing services, including AWS, GCP, Azure, OCI, Lambda, TensorDock, Vast.ai, RunPod and CUDO. It also can run workloads on on-prem servers.

The engine automates the complexities of setting up and running AI workloads so you can focus on data and research. And it can cut costs by taking advantage of cheap cloud GPUs.

dstack can be deployed in several ways, including open-source self-hosted, enterprise self-hosted and managed dstack Sky versions. The open-source version can be installed on personal cloud accounts or data centers, the enterprise version includes security and compliance features. dstack Sky is a managed service with a web console and marketplace for GPUs.

Some customers have found dstack useful, including:

• Convenience in setting up personal Slurm clusters with cheap cloud GPUs.
• Ability to define configurations in repositories and run them without worrying about GPU availability.
• Easy access to the best GPU options across multiple clouds, saving time and money.
• Combining the simplicity of Docker with the auto-scaling abilities of Kubernetes.

dstack has a lot of documentation and a community-run Discord channel for help and discussion. If you're interested in adopting AI technology, dstack can help you get there faster.

Published on June 14, 2024