http://github.com - automatic! GitHub
Pre-requisites :-> . kubeflow installed . tf-operator running in 'kubeflow' namespace . Single node setup as /tmp/results must be shared between
docker build . -t test/tf-job:v1 mkdir /tmp/tf-results kubectl create -f tf-job.yaml
This is a distributed tf training model with MASTER, WORKER & PARAMETER SERVER They need a shared storage, for now run all JOBs in single node. Will add NFS support soon.
Training model is parameterized and the params are taken as ENV variables in tfjob.yaml.
Trainig results will be avaialble in /tmp/tf-results directory of host.
kubectl get po -n kubeflow kubectl get tfjobs -n kubeflow kubectl get jobs -n kubeflow
========== Happy Training -> will add serving details soon ==========