Kubeflow 1.8: Machine Learning Platform Maturity and MLOps

Introduction

Kubeflow 1.8, released on July 18, 2023, is most relevant if you’re operating Kubernetes in production and want fewer surprises. This post walks through the highlights and the kinds of operational scenarios where the changes tend to matter first.

Pipeline Improvements

• Pipeline execution enhancements improve reliability and performance for complex ML pipelines.
• Component improvements provide better reusability and composability of pipeline components.
• Scheduling enhancements enable better resource allocation and job scheduling.
• Monitoring improvements provide better visibility into pipeline execution and performance.

Model Serving Enhancements

1. Inference performance optimizations reduce latency and improve throughput.
2. Multi-model serving enables serving multiple models from a single endpoint.
3. Auto-scaling improvements provide better scaling based on inference load.
4. A/B testing capabilities enable experimentation with different model versions.

MLOps Integration

• Version control integration enables better model versioning and tracking.
• CI/CD improvements provide automated model training and deployment pipelines.
• Monitoring enhancements enable better model performance and drift detection.
• Governance capabilities provide better model lifecycle management.

Developer Experience

• Notebooks improvements provide better Jupyter notebook integration and management.
• CLI enhancements simplify common ML workflows and operations.
• Documentation improvements provide better guides and examples.
• Templates expansion provides more ML pipeline templates.

Getting Started

kubectl apply -k github.com/kubeflow/manifests/kustomize/cluster-scoped-resources?ref=v1.8.0
kubectl apply -k github.com/kubeflow/manifests/kustomize/env/platform-agnostic?ref=v1.8.0

Create a Pipeline:

apiVersion: tekton.dev/v1beta1
kind: Pipeline
metadata:
  name: ml-pipeline
spec:
  tasks:
  - name: train
    taskRef:
      name: train-task
  - name: evaluate
    taskRef:
      name: evaluate-task
    runAfter:
    - train

Summary

Kubeflow 1.8 provides teams with powerful machine learning capabilities for building, training, and serving ML models on Kubernetes.