Updated 2021-05-17

Run TensorBoard Interactively

Overview

  • Running TensorBoard on the cluster is a multi step step process that is similar to running other interactive software:
    • Install Tensorflow into your Conda Env
    • Set up VNC Interactive Session
    • Set up tunnel for VNC viewer so you can access the interactive job
    • Load Tensorflow module (or install tensorflow into a conda environment, more details below)

Install TensorBoard into Conda Env

Important

  • Instead of installing TensorFlow into a conda environment, you can just use module avail tensorflow to list modules, and load a tensorflow module of your choice with module load tensorflow-gpu replacing tensorflow-gpu with whatever version you prefer. Loading the tensorflow module lets you use tensorboard without having to install it into a conda environment
  • tensorflow-gpu/2.0 is only supported on RHe7 queues

Warning

Before loading the anaconda module, make sure you have set up the correct symlinks for the .conda file. Follow the Conda Environment Guide for information to do so.

  • After you have set up the symlinks, you can either create a new conda env with Tensorflow or install Tensorflow into an exisiting environment
  • Load the Anaconda module with module load anaconda3/2019.10
  • Create a new Conda Environment that includes Tensorflow:
#Tensorflow CPU
conda create -n tensorflow_env tensorflow #replace "tensorflow_env with whatever you want to name the env

#Tensorflow GPU
conda create -n tensorflow_gpuenv tensorflow-gpu #replace tensorflow_gpu with whatever you want to name the env
  • To install in an exisiting env:
  • Use pip install or conda install
  • To Start env use source activate <your env name>
  • To quit using env run: source deactivate <your env name>
  • For more information on how to create, run, and use conda environments, check out the conda env guide

Setup VNC Session

  • Please see the VNC guide for instructions on how to set up the Interactive VNC session

Load Tensorflow Module

  • In the VNC window, open a terminal by clicking top left Applications > System Tools > scroll down to terminal. Once you scroll down to terminal you can also right click it to add it to the desktop and top bar for easy access

Tip

Once you scroll down to terminal in the drop down menu, you can also right click it to add it to the desktop and top bar for easy access

  • All commands here on will be typed in terminal in VNC
  • To see available versions of tensorflow run
module avail tensorflow
module load <version of tensorflow>  #Recommended: module load tensorflow-gpu/2.0 (only on RHe7)

Load Anaconda Module and Conda Env (if using TF in conda env)

  • If you are using tensorflow within a conda environment, follow these steps
  • In the VNC window, open a terminal by clicking top left Applications > System Tools > scroll down to terminal. Once you scroll down to terminal you can also right click it to add it to the desktop and top bar for easy access

Tip

Once you scroll down to terminal in the drop down menu, you can also right click it to add it to the desktop and top bar for easy access

  • All commands here on will be typed in terminal in VNC
  • To see available versions of anaconda run
module avail anaconda
  • Load anaconda module
module load <version of anaconda>  #Recommended: module load anaconda3/2019.10
  • Load your conda env that has Tensorflow with source activate <your conda env name>

LoadCondaEnv

Run TensorBoard

  • You can now run TensorBoard as you would normally
  • Run tensorboard --logdir=<your path to event logs>
  • Open firefox in the VNC Session by clicking on the firefox icon in the top bar
  • Paste the address generated by running tensorboard into firefox to go to the tensorboard site
  • Alternatively, you can just click the link in the terminal

Note

Important: Tensorboard might not render properly on RHEL6, the following screenshot was taken on RHEL7

Congratulations! You have successfully ran Tensorboard interactively on the cluster

LoadCondaEnv