Feature Area

/area sdk

As discussed in the Kubeflow SDK, we want to include the Kubeflow SDK in the python requirements of the KFP SDK.

This has several benefits:

1. Force a coherent product story for our end users.
2. Force the SDK to resolve package dependency issues earlier
3. Seed the Kubeflow SDK UI with the existing KFP UI
   • See this PR as a point of reference and this github issue
4. Force the SDK to prioritize migrating KFP

What feature would you like to see?

pip install kfp to include kubeflow

What is the use case or pain point?

From work that was done by Red Hat's @instructlab project (and likely other closed-source projects), the experience of using the Kubeflow Trainer with Kubeflow Pipelines is bad. The projects feel entirely separate even though they are meant to be used together.

Concretely, the code putting them together can be seen here: https://github.com/opendatahub-io/ilab-on-ocp/blob/main/training/components.py to get clarity on how challenging it is.

Is there a workaround currently?

Yes, but it's an extraordinary amount of code that I do not expect an end user to actually write.

Love this idea? Give it a 👍.