v0.8.10

• tensorflow: support for cholesky with upper arg, batched svd, and scipy.linalg.solve_triangular

Full Changelog: v0.8.9...v0.8.10

v0.8.9

• add swapaxes function for tensorflow

v0.8.8

Enhancements:

• autoray.lazy: add support for moveaxis and swapaxes

Full Changelog: v0.8.7...v0.8.8

v0.8.7

Enhancements

• Torch improvements:
  • Added torch implementation for trace with axes specified plus tests.
  • Allowed torch default_rng seed to be None.
  • Added axis kwarg support for torch count_nonzero.
• Added decorator support for registration APIs:
  • register_function can now be used as a decorator.
  • register_custom_wrapper can now be used as a decorator.
• Refactored internals to use the new decorator style for function registration.

v0.8.6

What's Changed

• add stop_gradient by @ChenAo-Phys in #29
• Add torch support for scipy.linalg.solve_triangular.
• Add tensorflow support for random.default_rng.
• Add allclose support for lazy arrays.
• Add Python 3.14 support.
• Fix and sort internal backend registration functions.

New Contributors

• @ChenAo-Phys made their first contribution in #29

Full Changelog: v0.8.5...v0.8.6

v0.8.4

Enhancements

• add autoray.is_scalar function
• lazy.einsum: for numpy backend set optimize=True by default
• lazy.linalg.norm: add kwargs support
• torch.random.default_rng: add choice support
• move ci to pixi

Full Changelog: v0.8.3...v0.8.4

v0.8.3

Enhancements

• register array and asarray as creation routines, so that they pick up device by default from like
• LazyArray.show: show more information about call signature by default
• lazy creation routines, support single int as 1D shape specifier
• remove custom torch.count_nonzero which torch now natively implements
• bump min python version to 3.10
• torch.take implementation: support vmapping over both scalar and sequence indices
• get_namespace: support scipy as a custom submodule.
• custom autograd.numpy.take to support grad (HIPS/autograd#743).

Full Changelog: v0.8.2...v0.8.3

v0.8.2

Ehancements

• Speedup creation of get_namespace.
• Reduce namespace attribute retrieval overhead to essentially zero

Full Changelog: v0.8.1...v0.8.2

v0.8.1

Bug fixes

• enable jax batched qr including for flat, grad inputs

v0.8.0

Breaking changes

• LazyArray.__iter__: now iterates over slices of array rather than the computational graph nodes

Enhancements

• add xp = autoray.get_namespace(like) as an alternative api
• LazyArray: add support for python array api via .__array_namespace__()
• LazyArray: support broadcasted linear algebra
• alias ("torch" "equal") to torch.eq
• lazy: support caching of more kwargs
• lazy: add take_along_axis
• lazy: add equal
• add "random.default_rng" implementation for jax and torch to support random pure functions (#27)

Bug fixes

• python compiler supports calls with kwargs
• jax: re wrap fat QR to support gradient
• pytensor: qr fix mode to 'economic'