PythonCall With Jax: Fast inference w/ numpy but does not work with jax.grad need jax.numpy, which is slow, for gradient

I need Jax for something that Zygote cannot do well (meta-learning) and someone recommended PythonCall as a solution to some issues I was having with PyCall.

So far, PythonCall has been great. Things work and it is generally quite quick.

There is one pain point: jax.grad does not work with numpy.array

Start by reproducing the reported difference between numpy.array and jax.numpy with PythonCall, focusing on the jax.grad entry point. Check the existing PythonCall interoperability behavior and record a minimal failing example. Done means jax.grad works with the numpy.array path without requiring the slower jax.numpy approach.

索引モデルが issue の本文から書いたものです。