Yuan Tang
@typhoonzero Why is `get_model()` needed? In ElasticDL, we can support both subclass and functional Keras models without this additional method.
Is there a contribution guide? How much of the maintenance efforts are manual vs. automated?
@sandeep-krishnamurthy Thanks for bringing this up. We are also tracking this in https://github.com/kubeflow/kubeflow/issues/4805.
Agreed. We should only expose modules to developers when needed. Otherwise it might introduce additional maintenance efforts such as backwards compatibility and versioning of the new module.
This indeed feel like unnecessary and complicate the implementations. Kubeflow/common should aim to be easy for downstream operators to implement.
What are the changes you are trying to make to training operator?
> > What are the changes you are trying to make to training operator? > > add some logic in pytorch job lifecycle, delete pods when job suspened, create pods...
Yes it's part of our roadmap so contribution is welcomed.
Yes they should be able to override the defaults.