Artifacts for installing the Distributed Workloads stack as part of ODH
Distributed Workloads is a simple, user-friendly abstraction for scaling, queuing and resource management of distributed AI/ML and Python workloads. It consists of the following components:
-
CodeFlare Operator to manage the control-plane components:
- Multi-Cluster Application Dispatcher (MCAD) for management of batch jobs
- Instascale for on-demand scaling of a Kubernetes cluster
-
CodeFlare SDK to define and control remote distributed compute jobs and infrastructure with any Python based environment
-
KubeRay for management of remote Ray clusters on Kubernetes for running distributed compute workloads
Integration of this stack into the Open Data Hub is owned by the Distributed Workloads Working Group. See this page for further details and how to get in touch.
| Component | Version |
|---|---|
| CodeFlare Operator | v1.0.0-rc.1 |
| Multi-Cluster App Dispatcher | v1.35.0 |
| CodeFlare-SDK | v0.10.1 |
| InstaScale | v0.0.9 |
| KubeRay | v0.6.0 |
Follow our quick start guide here to get up and running with Distributed Workloads on Open Data Hub.
For the V2 version of the ODH operator follow this guide instead.