OS MLOps
This repository contains a number of assets for implementing an open source MLOps approach using OpenShift Container Platform and Red Hat OpenShift AI (RHOAI).
- RHOAI Workbench Controller based on Kubeflow Notebook Controller for notebook-based development,
- Data Science Pipelines based on Kubeflow Pipelines for workflow orchestration and experiment tracking,
- RHOAI Model Serving based on KServe ModelMesh for model serving.
RHOAI Demo Pack
To quickly set up a RHOAI environment for fraud detection and object detection demos, follow these steps:
- (optional) Set up GPU enablement if GPUs are present in your cluster.
- Deploy the OpenShift Data Science operator on your OpenShift cluster. The demo pack has been tested with RHOAI 2.5.0.
- Deploy the OpenShift Pipelines operator. The demo pack has been tested with Pipelines versions 1.8 to 1.10.
- Deploy the OpenShift Serverless operator. The demo pack has been tested with Serverless version 1.31.0.
- Deploy the OpenShift Service Mesh operator. The demo pack has been tested with Service Mesh version 2.4.5-0.
- Clone this repository and navigate to
manifests
.
- Run
oc apply -f dependencies.yaml
. Wait until the DataScienceCluster
CR has been deployed.
- Run
oc apply -k .
Once the manifests have been deployed, your environment contains:
- A Minio instance as a lightweight S3 storage provider. You can manage the S3 buckets through the Minio UI through the
minio-ui
route URL in project minio
. Use minio
and minio123
for logging in.
- A Data Science Project
fraud-detection
for running the fraud detection demo. The pipeline server is instantiated and cluster storage and data connections are configured.
- A Data Science Project
object-detection
for running the object detection demo. The pipeline server is instantiated and cluster storage and data connections are configured. The OVMS model server is instantiatend for model deployment.
- A Data Science Project
ray-demo
for running the Ray demo.
- A number of community workbench images.
- A number of custom serving runtimes.
To get started with your demo, instantiate the respective workbenches.
Fraud detection
- In the RHOAI dashboard, enter the
fraud-detection
project.
- Create a new workbench with an arbitrary name and these parameters:
- image:
Trino
- existing cluster storage:
development
- existing data connection:
fraud-detection
- In the workbench, clone this repository, navigate to
notebooks/fraud-detection
and follow the instructions.
Object detection
- In the RHOAI dashboard, enter the
object-detection
project.
- Create a new workbench with an arbitrary name and these parameters:
- image:
Object detection
- existing cluster storage:
development
- existing data connection:
object-detection
- In the workbench, clone this repository, navigate to
notebooks/object-detection
and follow the instructions.
References