Patrick Kidger

So if you're asking in terms of JAX abstractions, then `jax.lax.custom_root` is also backed by a `custom_jvp` so they're very similar in that regard: https://github.com/jax-ml/jax/blob/349cbc24bfe716c91e8e6af6b12a9100766c9783/jax/_src/lax/control_flow/solves.py#L131-L132 If you're asking in terms...

I've not followed every aspect of the discussion here too closely, but I've certainly wanted to fix up the scale-invariance for a while. (And not dived into it.) Having scale-invariance...

Hi there! I'm not a pydantic user, so I'm not super familiar with the details of their API. But if I understand you correctly then I think: - As per...

So I think your example with dataclasses is succesfully raising an error because you have the initialisation of the dataclass within your try/except region, and it is this which fails....

Ok, so! A couple of different things going on here now. --- First of all, I'm actually a little bit surprised that this doesn't work already! When jaxtyping uses an...

Sounds reasonable to me! No reason to restrict to just beartype+typeguard. :)

I'm afraid this isn't something that can be modified in torchcubicspline. More generally I'm afraid this package is now only in maintenace mode, and won't be accepting new feature requests....

> if Google X ever wants to get deep into AI QA Haha! For the record I left there two years ago :) Part of a startup doing something completely...

Hmm, I'm not sure I understand this point. From the perspective of a top-level caller: if a nested hint 'partially succeeds' then this means that it fails, and so `state`...

Thanks for the report! (Though I'll note that your `google3` usage is not reproducible in the outside world ;) ) Trimming that out, the error ultimately is raised from this...