Train models on a pool of your own VMs from your laptop. All dependencies and artifacts are tracked.
Machine learning workflows consist of steps. Every step may have its own config, may depend on other steps, and may produce artifacts.
Define your workflows in the form of YAML files within your project.
No changes to your code are required.
Workflows run on a pool of machines that are called runners.
You can register any AWS' EC2 instance or GCP's or Azure's VM, or your own server as a runner by running a simple bash command.
Run any workflow via the CLI on your laptop. Override config values. Run a parameter sweep if you want to use a combination of multiple parameters.
The workflow will run on one of the available runners.
Because dstack is aware of your workflows, their exact steps, and input parameters, any run can be back-tracked and reproduced end-to-end on any infrastructure.
Sign up to join our waitlist