Question

Two ML teams at your company are independently prototyping models in Vertex AI Workbench notebooks. Leadership wants a single place where both teams can visualize training curves (loss, accuracy over steps), compare hyperparameter runs, and share interactive scalar and histogram plots with stakeholders who do not have notebook access. The teams already log metrics from their custom training loops. Which approach best satisfies these collaboration requirements with the least custom engineering?

Accepted Answer

Have each team log summaries with the TensorFlow summary writer and upload them to Vertex AI TensorBoard, then share the managed TensorBoard instance with stakeholders via IAM.

Answer

Export final metrics from each notebook to a shared BigQuery table and build a Looker Studio dashboard for stakeholders to view aggregated results.

Answer

Have each team run a local TensorBoard server in their notebook and periodically screenshot the plots into a shared Google Doc.

Answer

Store metric values as JSON files in a Cloud Storage bucket and grant stakeholders object viewer access to download and inspect them.