Justus Schock

@NielsRogge fair enough. Just wanted to get some confirmation that we actually need this and don't implement something again we already have on a different name :)

Unassigning for now. If it should be continued, feel free to reassign the new owner for this.

@tchaton is this still relevant or can this be closed? Research Studio does have databases, right?

It does not have an impact on the forward/backward passes in general, but some more complicated models use these hooks as part of their logic, because it is highly dynamical.

@awaelchli could we port this to lite?

Hey @zilto , thanks for the issue and sorry for the long time to reply. The code already looks pretty good to me. Regarding your questions: 1.) I'd do the...

calling @awaelchli for distributed review :)

Hey @ytang137 and sorry for the late reply. A metric takes the following input kwargs on initialization: - process_group: The process group on which the synchronization is called. Default is...

@SkafteNicki what's left here? :)

I'd probably name the things a bit differently and stick to string arguments only e.g no True and instead of False have a "none" arg the resulting docstring would then...