Integrate the Kubeflow Pipelines with Kubeflow SDK #125
Description
What you would like to be added?
The Kubeflow SDK goal is to provide unified interface to interact with Kubeflow tools, and we would like to create PipelinesClient() to give users ability to create ML pipelines, as we discussed at the last Pipelines WG.
Given that migration of kfp into kubeflow SDK might take time, @mprahl and @HumairAK proposed that we install kfp in the Kubeflow SDK packages, and create create a thin wrapper on top of the KFP Client:
class PipelinesClient(kfp.Client):
....Additionally, we still want to install Kubeflow SDK as part of kfp if users define it in their components: kubeflow/pipelines#12027. @Prateekbala has already opened PR here: kubeflow/pipelines#12368
I believe, that might help us to drive the integration forward and achieve the long-term goal of unified Python SDK to manage AI workloads on Kubernetes.
cc @kubeflow/wg-pipeline-leads @kubeflow/kubeflow-sdk-team @franciscojavierarceo @kubeflow/kubeflow-trainer-team
Why is this needed?
Kubeflow SDK should support Kubeflow Pipelines.
Love this feature?
Give it a 👍 We prioritize the features with most 👍