neilgirdhar / tjax Goto Github PK

Hey @NeilGirdhar !

I have a keen interest on having Pytree ecosystem in JAX as IMO it makes many things simpler without loosing too much generality.

I've created this library called Treeo which is somewhat similar to your dataclass implementation, while it also support dataclasses, it has some additional features.

Would like to know if you are interested in collaborating on these topics, I am specially interested in Pytree Optimizers that could integrate with Treex and Elegy. Happy to have a chat if you have the time.

Wait on python/mypy#14072

tjax requires numpy>=1.21 which seems to conflict with tensorflow's requirement of numpy~=1.19.2. I don't think needing both jax and tensorflow together is an uncommon use case. Jax functions can be converted to tensorflow (e.g., jax.experimental.jax2tf) which allows the use of, amongst other things, tensorflow's model saving/loading machinery.

This depends on python/mypy#14293.

Waiting on google-deepmind/tree#91 and re-release.

NEP 29 says to drop Python 3.8 on April 14, 2023. Use https://github.com/frostming/fix-future-annotations to upgrade the code.