This app is to simplify machine learning on Kubernetes for beginners. NOTE: This is a work in progress.
MVP of an end-to-end deployment app for machine learning jobs on Kubernetes clusters.
Enables single click installation of CUDA, CuDNN, Kubernetes components.
Single click bootstrapping of clusters uses kubeadm.
Single click activation of alpha features to enable discovery of GPUs.
Runs on Ubuntu machines.
No guarantee that everything works, use at your own risk.
Written in Go.
Future Work
Prometheus, Grafana dashboard for monitoring.
Tensorboard for visualizing training.
Support for custom schedulers.
Setup
Requires kubectl on the local machine
$ brew install kubectl
$ cd $GOPATH/src
$ mkdir -p github.com/swiftdiaries
$ cd github.com/swiftdiaries/
$ git clone github.com/swiftdiaries/dl-kops
$ cd dl-kops/src/app/frontend
$ export PORT_1=8081
$ go run dashboard.go