Named Tensors + jit


are there any plans to support jit compilation of code using named tensor operations?

IMHO, named tensors are potentially great, but are in a weird place: rewriting old code seems not worth it, especially with that list of unsupported subsystems. But even for new code, that is big and complex enough to require dimension reorderings, one has to choose between JIT and named tensors beforehand. To me, JIT seems more valuable (and even more so for simpler code fragments, that require no more that some occasional head/tail [un]squeeze).

Hi @googlebot,

We’ve discussed it but there are no immediate plans. However it wouldn’t be that hard to add. Maybe by the next release ?