Getting Started

Requirements

• Go >= 1.20
• Docker

Installation

When starting from scratch (you don’t have a kubernetes cluster) the quickest way to get up and running with Kai is to clone the repo and run the quickstart using make. This will create a cluster using kind, build and deploy the controller, and deploy an example pipeline.

Clone the repo and run the quickstart.

git clone https://github.com/dreamstax/kai && cd kai
make quickstart

Existing Clusters

note: this section is wip as we do not yet have a first release

Use kubectl to install the CRDs and deploy the controller to your cluster.

kubectl apply -f {github-release-url}

Usage

Create a new pipeline by defining a Pipeline resource.

# pipeline.yaml
apiVersion: core.kai.io/v1alpha1
kind: Pipeline
metadata:
  name: image-classifier
spec:
  steps:
  - spec:
      model:
        modelFormat: pytorch
        uri: gs://kfserving-examples/models/torchserve/image_classifier/v1

Then apply this pipeline resource to the cluster.

kubectl apply -f pipeline.yaml

Executing a Pipeline

note: this section is wip as we build out kai-piper

Potential steps…

• retrieve pipeline ID (pipeline resource could expose this, also available via kai-piper)
• port-forward kai-piper server (could also be registered on an ingress gateway)
• call /v1alpha1/pipelineJobs/{job_id}:run