TFX on Google Cloud AI Platform Pipelines: empty Tensorboard

I was able to complete the lab successfully. However, one of the optional exercises involved exploring the Tensorboard, which is reachable from the Kubeflow pipelines UI by clicking on the Trainer component.
Unfortunately, when I tried to launch the Tensorboard, it showed empty data even though the pipeline had already completed. That’s a bit unfortunate since I was curious to explore it.

Something that I noticed is that the lab package used TF version 2.3.4 and the latest available Tensorboard version, that could be launched from KFP UI was for TF 2.2.0.


Hi Ricardo! Yes this is an ongoing issue and we’ll follow it up with the Qwiklabs team. The lab is hosted on their end so we can’t troubleshoot it directly. It might indeed be a versioning issue because I’ve tried this lab before and Tensorboard worked out okay. An update may have broken it. Will let you know asap. Thank you!