Training operator 1.6 Roadmap

[johnugeorge]

Major features tracked for next Training Operator release. Few might be a stretch but still tracked for next release.

• Better KFP integration (PyTorchJob: making master/worker write to KFP artifacts #1580)
• PodGroup enhancements(trainning-operator may need to monitor PodGroup #1574)
• Integration with other training frameworks - Paddlepaddle(#520)
• Enhancements on Pytorch Elastic training (#1645, #1626)
• Support for multi platform image support (Support for multiplatform container imege (amd64 and arm64) #1664)
• Support for K8s 1.25 (Support Kubernetes v1.25 #1682)
• Implement conformance tests (Conformance tests for Training operator #1695)
• Support coscheduling plugin (Support coscheduling plugin #1722)
• [SDK] Create Unify Training Client([SDK] Create Unify Training Client #1719)

/cc @kubeflow/wg-training-leads

[terrytangyuan]

Looks great! Thanks for driving this!

[talcoh2x]

Hi, there is a plan to add kueue support for 1.6?

[tenzen-y]

Hi, there is a plan to add kueue support for 1.6?

@talcoh2x Probably, we can not support kueue in v1.6 since there are many tasks to support kueue and the feature freeze date is coming up.

Wednesday, Jan 25th 2023: Week 18 - Feature Freeze

[tenzen-y]

@johnugeorge Can you add #1645 #1722. Adding these helps to track new features for 1.6.

[tenzen-y]

Ah, #1645 has already been added. Could you add only #1722?

[johnugeorge]

@tenzen-y Done

[tenzen-y]

Great!

[andreyvelich]

Training Operator 1.6 has been officially released: https://github.com/kubeflow/training-operator/releases/tag/v1.6.0
Thank you very much for your contributions!