Using Kubernetes for Machine Learning Frameworks
summary
Explains why Kubernetes’ features (isolation, autoscaling, load balancing, GPU support) make it suitable for training and running ML workloads, and shows how to set up open-source frameworks—TensorFlow, Apache MXNet and PyTorch—on a Kubernetes cluster. Covers the steps for training, data preparation (“massaging”) and inference phases when deploying these ML frameworks on Kubernetes.
AI-written summary · about
chapters (26)
- 0:00Intro
- 1:24Centerpiece for digital transformation
- 3:54Machine Learning 101
- 6:26The Amazon ML stack: Broadest & deepest set of capabilities
- 11:33Amazon EKS-run Kubernetes in cloud
- 13:03Amazon EKS deployment
- 13:45Getting started with Amazon EKS
- 14:24Creating an EKS cluster using eksctl
- 15:22GPUs for Machine Learning training • Training maps to matrix multiplications • Coupled with extremely high memory bandwidth
- 17:08Set up K8s for ML-option 1
- 18:49Create K8s cluster for ML-option 1
- 19:20Scaling the cluster
- 23:18Create K8s cluster for ML-option 2
- 23:38Set up K8s for ML-option 2b
- 24:05Challenges in setting up containers for ML
- 25:24AWS deep learning containers
- 26:1616 container images
- 26:39ML on K8s—without KubeFlow
- 28:09MNIST database
- 28:59Fashion MNIST
- 29:59AWS is the platform of choice to run TensorFlow
- 31:16Machine Learning using TensorFlow on K8s
- 41:42Apache MXNet
- 47:43Distributed training using Horovod
- 49:22Machine Learning pipeline for K8s
- 50:07Machine Learning pipeline using SageMaker
related talks
31mKubernetes for Machine Learning: Managing and Scaling AI Workloads
42mKubernetes for Machine Learning: Managing and Scaling AI Workloads
26mMachine Learning Meets K8s
12mKeynote: Building Enterprise AI/ML Platform on Kubernetes
22mThe Containerization of Machine Learning: TensorFlow: Kubernetes, and Kubeflow
46mMachine Learning with Kubernetes
36mMulti-Cloud Machine Learning Data and Workflow with Kubernetes
27mHarnessing Kubernetes for Scalable AI/ML Workloads