View Vertex AI TensorBoard data

The Vertex AI TensorBoard web app provides a visualization of logs associated with a Vertex AI TensorBoard experiment. This web application offers several tools and dashboards to visualize and compare data across experiment runs, and to profile model training performance.

To learn more about each dashboard, see the README for "The Visualization".

Access a Vertex AI TensorBoard web app

The Vertex AI TensorBoard web app is accessed using the following address format:

https://REGION.tensorboard.googleusercontent.com/experiment/projects+PROJECT_NAME_OR_ID+locations+REGION+tensorboards+TENSORBOARD_ID+experiments+EXPERIMENT_NAME

• REGION: the region your TensorBoard is located in
• PROJECT_ID: the project your TensorBoard was created in
• TENSORBOARD_ID: the TensorBoard ID
• EXPERIMENT_NAME: the name of the TensorBoard experiment

This link can be shared and used by anyone who has permission.

You can navigate here directly by inserting the appropriate values, or by following a link available in the following ways.

gcloud CLI and Vertex AI SDK for Python

When creating or uploading logs to a Vertex AI TensorBoard, the gcloud CLI and SDK output a link to the Vertex AI TensorBoard instance.

Google Cloud console

You can access the TensorBoard web app using the Google Cloud console.

Experiments

If your Vertex AI TensorBoard is associated with a Vertex AI Experiment, which is automatic when a default instance is involved, you can access the TensorBoard web app directly from the Experiments page.

Custom training

If you used Vertex AI TensorBoard with custom training you can access the TensorBoard web app directly from the custom training job.

1. In the Vertex AI section of the Google Cloud console, go to the Training page. Go to Training
2. Click Custom Jobs tab.
3. Select your training job from the list of training runs. The details page appears
4. Click the Open TensorBoard button.
   

Vertex AI Pipelines

If you used Vertex AI TensorBoard with Vertex AI Pipelines, the TensorBoard web app can be accessed directly from the component information display.

1. In the Vertex AI section of the Google Cloud console, go to the Pipelines page. Go to Pipelines
2. From the pipelines tab, scroll or filter the pipeline list to find your pipeline.
3. On the pipeline display select the component where TensorBoard metrics are captured, typically the training component.
4. Select the Open TensorBoard button in the upper right.
   

Visualize Vertex AI TensorBoard metrics in the Google Cloud console

If your TensorBoard instance is associated with a Vertex AI Experiments, which is automatic when using a default TensorBoard instance, certain metrics can be visualized directly in the Google Cloud console, without using the web application.

The Compare and analyze runs document describes how to view details of your experiment runs and compare the experiment runs to each other in the Google Cloud console. TensorBoard timeseries metrics appear under the Charts tab in the navigation bar of the Vertex AI Experiments console.