Developer platform for reproducible AI

Define deep learning workflows. Run them via the CLI. Get resources provisioned automatically. Track data and model lineage.

Error. Your form has not been submittedEmoji
This is what the server says:
There must be an @ at the beginning.
I will retry
Reply
Free forever. No credit card required.
Dstack website hero
Device frame

Prepare. Train. Finetune. Reproduce.

Codify your workflows and the resources they need. Run workflows and keep track of logs and artifacts in real-time.

  • Workflows

    Define your workflows and the resources they need via a simple and extensible YAML format.

  • Runners

    Use your own servers as runners, or let dstack set up runners in your own cloud account based on the need.

  • Tools

    Run workflows and watch their progress via the CLI. Browse runs, logs, and artifacts via a convenient dashboard.

  • Logs

    Browse workflow logs in real-time as your workflow is running.

  • Artifacts

    Persist output artifacts in S3-compatible storage in real-time as your workflow is running.

  • Tags

    Tag successful runs with tags for later reuse.

Define reproducible workflows

Define every step of your training process as an individual reproducible workflow. 

For each workflow, define variables, specify a Docker image, commands, other workflows it depends on, and output artifacts.

No changes in your code are required.

Dstack website zoom workflows

Run and manage workflows from the CLI

Run workflows via the CLI. Browse real-time logs and artifacts. Stop and resume runs.

Mark successful runs with tags. Reuse them yourself and share them with others.

Dstack website zoom run

Host runners on your own servers

Register your local machine or any remote servers as self-hosted runners.

After you set up self-hosted runners, submit any workflow to run there from your laptop using the CLI.

Dstack website zoom self hosted

Connect to your cloud account

Authorize dstack to run workflows in your own cloud account. 

Use either regular or spot instances. 

If a spot instance terminates before finishing the run, dstack will re-assign it to another instance without losing the checkpoints.

Dstack website zoom autoscale 1

Can be used together with any framework, tool, or cloud vendor

🙌

Register to get an invite to the early preview program

Error. Your form has not been submittedEmoji
This is what the server says:
There must be an @ at the beginning.
I will retry
Reply
Free forever. No credit card required.
Runs on Unicorn Platform