Tensorboard

From UFRC
Revision as of 21:03, 12 December 2023 by Magitz (talk | contribs)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigation Jump to search

Back to TensorFlow page

TensorBoard is used for visually monitoring statistics while a neural network trains.

Our recommendation for using TensorBoard on HiPerGator is to run TensorBoard monitoring as a separate workload within a HiPerGator Desktop session. Many neural network software frameworks are able to stream training data to a TensorBoard log file.

Here are the steps to monitor the log separately from your training workload

  • Request a HiPerGator Desktop via the Interactive Apps Menu of OnDemand
    • For resources, something like 1 CPU and 2 GB RAM should be good.
  • Connect to the session when it starts and open a "Terminal Emulator" from the desktop Applications menu in the top left.
  • Start a TensorBoard instance in the resulting terminal/shell (replace "./pathtolog" with the path to where your TensorFlow process is storing its logs):
  $ module load tensorflow ubuntu
  $ tensorboard --logdir ./pathtolog &
  • After a few seconds, TensorBoard will display a message similar to this (There may be some errors displayed, but TensorBoard should still work):
...
TensorBoard 2.4.0 at http://localhost:6006/ (Press CTRL-C to quit)
  • Copy the printed URL ("http://localhost:6006/" in the example above)
  • Press <ENTER> to get back the '$' prompt
  • Start a browser:
$ firefox (or chrome)
  • Paste the TensorBoard URL into the browser's location bar to view the TensorBoard UI.