Albert Steppi

Closing to allow resync after #777

> Regarding point 5: @steppi, would it be nice to add a default style for the buttons downstream to `jupyterlite-sphinx` that can claim compatibility with PST, i.e., to update/deprecate `:example_class:`...

I can clean up the merge conflicts and will take a close look at everything.

I've resolved the merge conflicts and carried on from there to add dispatch to all available special functions across backends. I'll post a comment explaining everything tomorrow. I needed to...

I'll investigate the failures and try to figure out why they aren't happening locally.

Hopefully all of the tests will pass now. The issues 1. I didn't realize we need to set `SCIPY_DEVICE="cuda"` to get the torch tests to run on GPU. 2. `betaln`...

The JAX GPU tests are failing because the JAX in CI is `v0.6.0`, but `hyp2f1` landed in `v0.6.1`. For whatever reason, `v0.6.1` and `v0.6.2` are up on PyPI but haven't...

``` =================================== FAILURES =================================== ______________________________ test_pickle[cbrt] _______________________________ func = @pytest.mark.parametrize("func", [nfo.wrapper for nfo in _special_funcs]) def test_pickle(func): > roundtrip = pickle.loads(pickle.dumps(func)) ^^^^^^^^^^^^^^^^^^ func = scipy/special/tests/test_support_alternative_backends.py:294: _ _ _ _ _...

Thanks @lucascolley. It looks like the failures are unrelated so far. A few comments I wanted to add. 1. I didn't add `special.sinc` because it isn't in `jax.scipy.special` but instead...

Crossing fingers, but that should take care of the last of the failures. On `torch`, when the default dtype is set to `float32`, the function `_skip_or_tweak_alternative_backends` was converting integer arguments...