It would be great to see the final results right away after the model was trained. We can copy the final-eval Taskcluster task to run after the export and upload tasks have completed in the main training pipeline.
Alternatively, we could run it on Cron, but it uses a GPU task, so it would be a waste of resources as we don't train models often.