This PR introduces a few reworkings of some of our basic numba algorithms:
So far:
[x] 1.8x faster implementation of vanilla
[x] corresponding vanilla_vjp (only 1.2x faster)
[x] 40x faster (🚀!) version of vanilla_stable
[x] 10x faster version of vanilla_stable_batched
Also renames the setting from USE_VANILLA_AVERAGE to STABLE_FOCK_CONVERSION (much clearer)
This PR introduces a few reworkings of some of our basic numba algorithms:
So far: [x] 1.8x faster implementation of
vanilla
[x] correspondingvanilla_vjp
(only 1.2x faster) [x] 40x faster (🚀!) version ofvanilla_stable
[x] 10x faster version ofvanilla_stable_batched
Also renames the setting from USE_VANILLA_AVERAGE to STABLE_FOCK_CONVERSION (much clearer)