Dilshod Tadjibaev

> From my point of view, this should be an either or: > > - Either, all operations track their shapes, potentially duplicating some code as well as possibly having...

> > We don't need to track the output shapes statically because the Burn API tracks and calculates dynamically. Burn needs only rank output. > > Indeed, Burn doesn't need...

Another example why should get away from static input shapes: https://github.com/tracel-ai/burn/pull/2441#pullrequestreview-2405432778

Closing this PR in favor of https://github.com/tracel-ai/burn/issues/2478

> Just a minor comment for now. > > For the new shape type introduced this LGTM. > > I'd like to get @antimora's feedback though. I know you wanted...

It seems this issue is outdated and don't have access to the file to recreate. Closing for now.

Linking #1689

Yes, I agree we should prevent 0d tensors and instead create Scalar type outlined in #1689.

Burn project's Ndarray backend would very much appreciate if this PR is merged. We have a long outstanding issue: https://github.com/tracel-ai/burn/issues/302

Burn is looking forward to this feature. It seems it's close =D