Limitations
The Compute API and the AI Factory platform are in their pilot phase and represent an MVP, not a finished product. Major changes can happen at any time, including changes to core concepts.
This page lists what does not work yet or works only in part. We update it as the platform changes, so check it for the current state.
Compute API
The Compute API is the REST API for deploying and managing services and models on the platform. It is listed in the infrastructure section of the API Catalog. It currently consists of Slurm, Kubernetes, and Harbor endpoints. The full reference is in the Swagger UI.
Slurm endpoints
The Slurm endpoints are currently unavailable for test users. They need access to Slurm, and only the Datalab team grants this access at the moment. We are working on a way for you to connect your own Slurm cluster, so that you can use these endpoints with external clusters.
Kubernetes endpoints
Call the provision endpoint once before you use any other Kubernetes
endpoint. The endpoints exist, but capacity is currently limited. Because of
this, you cannot deploy a usable model with the create_inferenceservice
endpoint at the moment. Therefore, the deploy_adapter endpoint works only
for platform models. For now, please use the service and stateful service
endpoints for CPU workloads.
Harbor endpoints
The Harbor endpoints support POST and GET. Call the provision endpoint
first.
Secrets
Do not pass secrets, for example for the stateful service endpoints, as environment variables through the endpoints. If your service needs secrets, include them in an image in your private Harbor project for now. Do not push that image to a public registry. Secret handling in the platform is currently in development.
Questions
If you have questions or concerns, contact us at software.platform@ai-at.eu.