Machine Learning using Kubeflow and Kubernetes
summary
Shows a reproducible path to run Kubeflow on Kubernetes with concrete tools and numbers: Kubeflow on EKS with a 8 GPU / 64 CPU node example, MiniKF for desktop, Fairing to package notebooks into ECR images, and Argo-backed pipelines that compile Python to YAML. Covers hyperparameter tuning with KFServing and Katib, node-group patterns (dedicated vs unified clusters), Cluster Autoscaler vs Escalator for batch ML, Horovod for distributed training, and a Credibly case that runs in production and benchmark configs that can be copied.
AI-written summary · about
chapters (20)
- 0:00Intro
- 1:13Machine Learning is Hard!
- 5:44Storage and Analytics for Machine Learning
- 11:21Amazon EKS: run Kubernetes in cloud
- 12:20Getting started with Amazon EKS
- 12:56Set up K8s for ML: Option 1
- 17:04Scaling the cluster
- 20:06Kubeflow Requirements
- 21:04Kubeflow on Desktop
- 22:16Kubeflow on Cloud
- 30:30Jupyter Notebook
- 31:23Kubeflow Fairing
- 35:04Hyperparameter Tuning using Katib
- 37:27Katib System Architecture
- 40:51Pluggable Interface
- 41:43Distributed Training using Horovod
- 43:29Kubeflow Pipelines
- 45:48Creating Kubeflow Pipeline Components
- 46:40Consumer Loan Acceptance Scoring
- 49:44Machine Learning pipeline for kubernetes on AWS
related talks
54mScaling Machine Learning Pipelines in Cloud
10mKeynote: Machine Learning using Kubeflow and Kubernetes
14mKubeflow: Where ML Meets Kubernetes
46mDeploying scalable Machine Learning models with Kubernetes
30mKubeflow, Your ML Tools on Kubernetes
39mDeploy your ML Workloads on Kubernetes with Kubeflow
28mMachine Learning on Kubernetes at Shell: A Kubeflow Journey
31mKubeflow - Data Science on Steroids