Dive into Kubeflow

Presented 5/10/2018 and Cloud Native Easy Bay Meetup at Adobe

Transcript

2. OpenAI Scaling Kubernetes to 2,500 Nodes https://blog.openai.com/scaling-kubernetes-to-2500-nodes/
4. Foundations Containers, Kubernetes Why? 1. Typical ML development workflow and some of its shortcomings 2. How can a distributed system like Kubernetes help us improve this flow? Labs aka.ms/kubeflow-labs
5. How? Setup a Kubernetes cluster (acs-engine/AKS) Demos 1. Running a basic Docker container 2. Running a TensorFlow job with GPU 3. JupyterHub 4. Distributed TensorFlow 5. Hyper-parameter sweeping 6. TensorFlow Serving
9. https://github.com/Azure/kubeflow-labs/tree/master/1-docker
10. OpenAI - Building the Infrastructure that Powers the Future of AI
12. Azure/acs-engine github.com/kubernetes/autoscaler/tree/master/cluster-autoscaler/cloudprovider/azure https://docs.microsoft.com/en-us/azure/aks/gpu-cluster
13. https://github.com/Azure/kubeflow-labs/tree/master/2-kubernetes
14. • Kubeflow project is dedicated to making deployments of machine learning workflows on Kubernetes simple, portable and scalable. • https://github.com/kubeflow/kubeflow
15. https://github.com/Azure/kubeflow-labs/tree/master/4-kubeflow https://github.com/Azure/kubeflow-labs/tree/master/6-tfjob
16. • multi-user Hub • spawns, manages, and proxies multiple instances of the single-user Jupyter notebook server.
17. https://github.com/Azure/kubeflow-labs/tree/master/5-jupyterhub
18. !
19. https://github.com/Azure/kubeflow-labs/tree/master/7-distributed-tensorflow
21. Andrej Karpathy's Image painting demo
22. https://github.com/Azure/kubeflow-labs/tree/master/8-hyperparam-sweep
23. • Provides out-of-the-box integration with TensorFlow models • Multiple models (or multiple versions of the same model) can be served simultaneously
24. https://github.com/Azure/kubeflow-labs/tree/master/9-serving
25. one) Solution is Kubernetes: • Highly Scalable • Easy to explore hyper-parameters space • Easy to do distributed training But really, Data Scientists shouldn’t have to care about containers, kubernetes and all that stuff • Pachyderm can version datasets and trigger new trainings when changes occur • Distributed File Systems • NFS • HDFS • … Classic DevOps solutions: • Containers • CI/CD • Autoscaling • A/B testing and canary release of Models • Comparing Production accuracy vs expected accuracy when possible • Rolling-updates • …
26. aka.ms/kubeflow-labs