TensorBoard is a visualization tool from the TensorFlow team. It’s a web based application that lets you view TensorFlow event logs, which contain a variety of useful information associated with a run:

  • Metrics (scalars)
  • Images, audio, and text generated during training
  • Model graph
  • Model statistics

For more information, see TensorBoard: Visualizing Learning.

TensorBoard and Guild

Guild integrates TensorBoard to make it easy to visualize TensorFlow event logs. To visualize events for a set of runs, you can launch TensorBoard by running:

guild tensorboard

You can also filter runs using Guild’s run filtering options. For example, to view only runs for the train operation in TensorBoard, use:

guild tensorboard --operation train

For more information, see tensorboard.

TensorBoard is also integrated with Guild View. You can launch TensorBoard from Guild View by clicking View in
TensorBoard which is located in the upper left of the screen.

Run synchronization

When you run TensorBoard from Guild, by either the tensorboard command or from Guild View, the list of runs is automatically synchronized with the current run view.