This document explains how to setup TensorBoard on Amazon EKS.
- Create EKS cluster using GPU
To set up TensorBoard one has to have event file of training available to see the visualization or summary of it. For completeness I have added a program (tensorboard/tbdemo.py
) which performs addition and write the summary to event file. I have also added a Dockerfile (tensorboard/Dockerfile
) which has tensorboard as well as event file generated by tbdemo.py
. To complete the experiement a docker image (rgaut/tensorboard
)has been publish.
-
File
tensorboard/tensorboard_demo.yaml
creates a deployment which runs the tensorboard on event files. A service is also being created so that user can access tensorboard via browserkubectl create -f tensorboard/tensorboard_demo.py
Note: Some time you might have to do port-forwarding as well in case if you are under the firewall network by running below command.
kubectl port-forward svc/tensorboard 6006:6006
Once the deployments and services are up goto your browser and type
127.0.0.1:6006
to access the tensorboard visusalization.