Skip to main content

Introduction

Substratus is a cross-cloud substrate for training and serving ML models. Substratus extends the Kubernetes control plane to orchestrate ML operations through the addition of new API endpoints: Model, Server, Dataset, and Notebook.

Why Substratus?​

  • Fine-tune and deploy OSS LLMs with no code required and built-in optimizations.
  • Launch remote Jupyter notebooks with a single command.
  • Run anywhere Kubernetes runs.
  • Install with minimal dependencies (a bucket and container registry).
  • Keep your data inside your network.
  • Scale to zero (save on GPU costs).
  • Create reproducible environments using containers.
  • Leverage GitOps out of the box.

Intro Video (under 2 minutes)​

Next steps​

Learn more in the Overview section or jump straight into the Quickstart guide.

Don't forget to star the GitHub repo and request the features you would like to see!

Note: Substratus might still introduce backward incompatible changes to the APIs until v1.0 is released.