Highlights

• xpx.testing.lazy_xp_function with jax_jit=True now handles functions which accept and return arbitrary wrappers around JAX arrays, by treating all arguments and return types that are not JAX arrays as static.
  • the static_argnums and static_argnames parameters are deprecated in favour of this.
• xpx.one_hot: new function to create one-hot vector arrays
• xpx.default_dtype: a convenience function for common queries to xp.__array_namespace_info__().default_dtypes()
• support for PyTorch device='meta'

Progress has also been made towards exposing functions like assert_equal under xpx.testing, but this API is still private for now.

What's Changed

• DEV: bump version to 0.8.0.dev0 by @lucascolley in #292
• ENH/TST: xp_assert_ enhancements by @mdhaber in #267
• MAINT: test against Dask 2025.5 by @crusaderky in #295
• MAINT: reshape does not accept int by @crusaderky in #294
• MAINT: upgrade to array-api-compat >=1.12 by @crusaderky in #296
• ENH: jax_autojit by @crusaderky in #284
• DOC: autojit notes by @crusaderky in #297
• TST: rework tests for xp_assert_equal by @crusaderky in #301
• ENH: support PyTorch device='meta' by @crusaderky in #300
• ENH: New function default_dtype by @crusaderky in #310
• ENH: add new function one_hot by @NeilGirdhar in #306
• REL: prepare v0.8.0 release by @lucascolley in #312

New Contributors

• @mdhaber made their first PR in #267

Full Changelog: v0.7.2...v0.8.0