Jon Harmon

Oh, duh, no, `keras::text_tokenizer` doesn't easily do the wordpiece stuff. Check out `wordpiece_encode` in https://github.com/bnosac/sentencepiece though to see if that looks efficient.

It isn't just class, anything else the provided dataset has (in my case we have some methods) should come through. After implementing a workaround, I think the best option might...

Yeah, I figured there might be some... massaging of that language. And the CONTRIBUTING.md would lay out exactly what's welcomed vs what might require a discussion, in any case. Probably...

Is there a boilerplate that should be added to `cran-comments.md` for this, or should we ignore it completely?

@gaborcsardi thanks! I *thought* that was the case but wasn't sure, and was led back to this issue by someone else's cran-comment.md mentioning it

Well, I *didn't* care about this, but now I definitely do! Realistically, though, I can imagine running something like https://twitter.com/RBERTbot on my Raspberry Pi, so I can definitely imagine wanting...

The description of the [Superseded](https://lifecycle.r-lib.org/articles/stages.html#superseded) label might make things a bit tricky/confusing: > Superseded functions will not receive new features, but will receive any critical bug fixes needed to keep...

I feel like I want the ability to do this as long as it doesn't make things harder to use. At first it was just "this feels useful," but I...

Note: This would be exported and usable within a UI, but doesn't impact the automatic "choose a UI" functionality.

Are you open to PRs to implement these? It would be a learning effort for me (working with cpp) but one I've wanted to take on.