Dan Foreman-Mackey

Thanks for the update - sounds good! But, like I've mentioned several times now, please don't plan on just revising in one go. It's _much_ more useful to _iterate_ on...

Thanks for looking into this! I'm seeing this on my macbook with commit `44b16f48cb5c5e7b9f5ed2b4d6a14ed14ab302e1` of Eigen (the current tip of the `master` from https://gitlab.com/libeigen/eigen). The original error showed up in...

Bisecting was actually fast! It looks like the offending Eigen commit actually is [`2d4c9b400cca33d2f5cf316efc7151236244edb1`](https://gitlab.com/libeigen/eigen/-/commit/2d4c9b400cca33d2f5cf316efc7151236244edb1). I'm happy to open a PR using this Eigen commit to try to reproduce in CI...

I expect @hawkinsp is right here! I've opened https://github.com/wjakob/nanobind/pull/749 demonstrating the issue in CI, and everything works fine using GCC on Ubuntu, but fails on macOS with clang. Hopefully this...

What do you mean by "respect the in_axes and out_axes"? It might make more sense to have this conversation in https://github.com/google/jax/issues/23624 regardless. If the example code I provided over there...

Closing in favor of: https://github.com/jax-ml/jax/pull/23881

Thanks for this request! This is something we'd love to support. We don't have a specific timeline, but I wanted to just confirm here that this feature request is acknowledged.

Adding one clarification: jaxlib doesn't actually come with its own LAPACK library. It actually links to the one from scipy. The code used to populate our API with those symbols...

Glad you got that figured out! Yeah, our plan is to migrate all the custom calls to the FFI in the near future (see https://github.com/google/jax/issues/23056), so in the long run,...

Sure! The place where this is specified on the JAX side is via the `operand_layouts` and `result_layouts` parameters in the custom call lowering. For a `n` dimensional input, we pass:...