Kubernetes for Machine Learning: Managing and Scaling AI Workloads
summary
Walks through a concrete MNIST pipeline built with Kubeflow: local Jupyter experiments, TensorFlow CNN training deployed as a TFJob CRD on Kubernetes, training on GKE (or AKS) with GPU nodes, serving via TensorFlow on GKE and monitoring with Stackdriver. Shows the exact kubectl/YAML patterns used, how to name pods and read TFJob events and logs, and practical scaling rules using node affinity and pod anti-affinity with fallback pods, plus why continuous hyperparameter tuning and staged load tests are required for production.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 1:07Why combine Kubernetes with machine learning
- 3:18Kubeflow for ML
- 4:41ML workflow experiments
- 7:10ML models to production
- 11:03MNIST with TF/Kubeflow
- 13:32TF jobs on Kubernetes
- 17:07ML lifecycle with Kubeflow
- 19:35Scaling ML workloads
- 34:22Resources for learning
- 35:25Q&A
related talks
31mKubernetes for Machine Learning: Managing and Scaling AI Workloads
46mDeploying scalable Machine Learning models with Kubernetes
22mThe Containerization of Machine Learning: TensorFlow: Kubernetes, and Kubeflow
51mMachine Learning using Kubeflow and Kubernetes
54mEconomics and Best Practices of Running AI/ML Workloads on Kubernetes
26mMachine Learning Meets K8s
46mAccelerating Machine Learning DevOps with Kubeflow
10mKeynote: Machine Learning using Kubeflow and Kubernetes