Tensorboard support about kubeflow HOT 10 CLOSED

There will be any coordination with tensorflow/tensorboard#92?

from kubeflow.

from kubeflow.

Still not cause I think that you are still on an internal planning phase so it is hard to avoid overlaps. But I hope that the plan and the internal roadmap will be made public soon with contribution welcome labels.

from kubeflow.

from kubeflow.

I meant on TB side. It is what seems by tensorflow/tensorboard#92 (comment)

from kubeflow.

Yes. We talk to @jart regularly and are trying to figure out how to bring TB as a service to Kubeflow.

from kubeflow.

We're very interested in bringing TensorBoard to Kubeflow. Doing this has raised challenging concerns that caused the TensorFlow team to think more about how it logs and loads data.

Right now we're focusing on taking an openly portable first approach. We want to get a rudimentary SQLite data model off the ground to learn more about how we can meet our requirements. You can see some of the work that's been done in summary_db_writer.cc and schema.cc.

While ubiquity is of the utmost concern, we have some confidence at this point that SQLite can carry a respectable amount of data when used in conjunction with reservoir sampling and an old school engineering mindset. Most importantly it offers big gains in terms of latency and accessibility.

While I can't speak for Kubeflow, I know SQLite can work in Kubernetes environments. While no single DB can meet the requirements of every organization, please note SQLite is but an appetizer. There are many other possibilities we're looking forward to exploring.

from kubeflow.

Duplicate of #297

from kubeflow.

Tensorboard is not comming up on kubeflow
put this in root /mlpipeline-ui-metadata.json
{
"version": 1,
"outputs": [
{
"type": "tensorboard",
"source": "/mnt/share/tflogs"
}
]
}

from kubeflow.

Any news regarding tensorboard in KF? We are using KF on GCE, I see it deployed even a TensorBoard pod, but cannot find any straight-forward way how to incorporate it into fairing.

from kubeflow.