This PR introduces two new components: A Kubeflow orchestrator and a Kserve model deployer.
Welcome Kubeflow
The implementation for Kubeflow in the context of this PR is done on a GKE cluster but it can easily be replicated to other clouds and we'll see that in the form of AWS and Azure recipes with Kubeflow support in the future.
Kserve Module
With this PR, there's also a brand-new in-house module - Kserve. A serverless installation of Kserve is implemented as a module. Components used in this installation are:
An Istio networking layer
A cert manager
A Knative installation
And finally, Kserve 0.9.0!
This module can be used across multiple recipes but the only customization available as of now is the namespace for deploying the kserve workloads. This limitation is due in part to the use of standard YAML manifests for the installation process. Support for any speicifc modification would mean changing the manifests and feeding to the kubectl function and we might see this in the future.
This PR introduces two new components: A Kubeflow orchestrator and a Kserve model deployer.
Welcome Kubeflow
The implementation for Kubeflow in the context of this PR is done on a GKE cluster but it can easily be replicated to other clouds and we'll see that in the form of AWS and Azure recipes with Kubeflow support in the future.
Kserve Module
With this PR, there's also a brand-new in-house module - Kserve. A serverless installation of Kserve is implemented as a module. Components used in this installation are:
This module can be used across multiple recipes but the only customization available as of now is the namespace for deploying the kserve workloads. This limitation is due in part to the use of standard YAML manifests for the installation process. Support for any speicifc modification would mean changing the manifests and feeding to the
kubectl
function and we might see this in the future.