Search Criteria
Package Details: python-jax 0.5.3-1
Package Actions
Git Clone URL: | https://aur.archlinux.org/python-jax.git (read-only, click to copy) |
---|---|
Package Base: | python-jax |
Description: | Composable transformations of Python+NumPy programs |
Upstream URL: | https://github.com/jax-ml/jax/ |
Keywords: | autograd deep-learning jax jit machine-learning ml xla |
Licenses: | Apache-2.0 |
Groups: | jax |
Submitter: | daskol |
Maintainer: | daskol (ei-grad) |
Last Packager: | daskol |
Votes: | 14 |
Popularity: | 1.47 |
First Submitted: | 2019-08-04 14:29 (UTC) |
Last Updated: | 2025-03-21 23:07 (UTC) |
Dependencies (12)
- python-absl
- python-jaxlibAUR (python-jax-rocmAUR, python-jaxlib-binAUR, python-jaxlib-cudaAUR)
- python-ml-dtypes
- python-numpy (python-numpy-gitAUR, python-numpy1AUR, python-numpy-mkl-binAUR, python-numpy-mkl-tbbAUR, python-numpy-mklAUR)
- python-opt_einsum
- python-scipy (python-scipy-gitAUR, python-scipy-mkl-binAUR, python-scipy-mklAUR, python-scipy-mkl-tbbAUR)
- python-build (make)
- python-installer (make)
- python-setuptools (make)
- python-wheel (make)
- python-protobuf (python-protobuf-gitAUR) (optional) – Australis
- python-requests (optional) – Distributed computations
Required by (39)
- mitsuba3-git
- python-array-api-compat (optional)
- python-autoray (check)
- python-blackjax
- python-chex
- python-d2l (optional)
- python-dm-haiku
- python-eagerpy-git (optional)
- python-equinox
- python-etils (optional)
- python-felupe (optional)
- python-finitediffx
- python-flax
- python-galpy (optional)
- python-jax-jumpy (check)
- python-jax-jumpy (optional)
- python-jaxtyping (optional)
- python-jmp
- python-kernex
- python-mpi4jax
- Show 19 more...
Latest Comments
« First ‹ Previous 1 2 3 4
daskol commented on 2021-02-01 15:04 (UTC)
Well, the issue is binary compatibility between
tensorflow
andjax
sincejaxlib
heavily use code located intensorflow
source tree. Now I have figured out that these distributions are independent enough or there are at lease guarantees of independence (but there are still some risks) so we can buildjaxlib
from sources.I don't think that jaxlib is intended to use. I guess that the reason is that
jaxlib
is a private thin bridge between XLA client attensorflow
source tree andjax
. This bridge also contains XLA extensions like PRNG. Nevertheless,jaxlib
andjax
like seemingly should be different packages since they have different release cycles and say some critical hotfixes could come tojax
but notjaxlib
. So, I'll addpython-jaxlib
AUR package in near future.edh commented on 2021-01-12 13:46 (UTC) (edited on 2021-01-12 13:46 (UTC) by edh)
The packages are designed to be two separate entities. PyPi tells you so (two distinct setup.py files), the developers treat them as such and jaxlib can in theory be used independently from jax. On a very basic level, jaxlib simply wraps XLA code while jax is a very fancy numpy wrapper.
If you want, I can take over maintainership of the package if you no longer actively use it.
P.S. the package is still hopelessly outdated. Why did you unflag it?
daskol commented on 2021-01-12 12:19 (UTC) (edited on 2021-01-12 12:19 (UTC) by daskol)
Sure. I have made the package the long time ago when build from sources were much more tricky.
Why?
edh commented on 2021-01-05 14:21 (UTC)
Can you please split the package into jaxlib and jax. Furthermore, it would be awesome if you would build the package from source the way it is described in the installation section at https://jax.readthedocs.io/en/latest/developer.html#building-from-source .
« First ‹ Previous 1 2 3 4