For Macs with >= 8 performance cores, we select CPU+GPU (original
attention). Otherwise we select CPU+ANE (split einsum).
Some computers (M1 Pro, 16 core GPU) might yield slightly better
performance using CPU+GPU+ANE with SPLIT_EINSUM.
Code signing is disabled for macOS. Users are instructed to update the
development team when they fork or clone the repo, if they want to
target iOS.
Co-authored-by: Fahim Farook <fahimf@gmail.com>
Pedro Cuenca