Package Details: python-jaxlib 0.4.35-1

Git Clone URL: https://aur.archlinux.org/python-jaxlib.git (read-only, click to copy)
Package Base: python-jaxlib
Description: XLA library for JAX
Upstream URL: https://github.com/jax-ml/jax/
Keywords: deep-learning google jax machine-learning xla
Licenses: Apache
Groups: jax
Submitter: daskol
Maintainer: daskol
Last Packager: daskol
Votes: 7
Popularity: 0.51
First Submitted: 2021-01-12 12:50 (UTC)
Last Updated: 2024-10-23 00:00 (UTC)

Latest Comments

« First ‹ Previous 1 2 3 4 5 6

daskol commented on 2021-11-12 12:07 (UTC) (edited on 2021-11-12 12:07 (UTC) by daskol)

[community/bazel] is already at 4.2

@firoz Right. JAX team pins bazel version to ensure reproducibility of builds (see .bazelversion). So, we should use exactly 4.1.0. The message you see should be ignored since it appears during bazel version check.

Does it build python-jaxlib for you?

firoz commented on 2021-11-12 11:00 (UTC)

ERROR: The project you're trying to build requires Bazel 4.1.0

https://archlinux.org/packages/community/x86_64/bazel/ is already at 4.2

steventrouble commented on 2021-03-09 00:55 (UTC) (edited on 2021-03-09 00:56 (UTC) by steventrouble)

Researchers: If you're having trouble getting CUDA to work, just make your own PKGBUILD that uses pip rather than trying to debug this or build from source. Trust me, it'll be faster and safer than trying to use pip directly.

To get it to work, you can download this PKGBUILD and modify the "pip install" line to run the pip install line from the jax github page:

package() {
    pip install \
        --ignore-requires-python \
        --root=$pkgdir \
        "jax" \
        "jaxlib==0.1.61+cuda112" \
        -f https://storage.googleapis.com/jax-releases/jax_releases.html
}

In general, this works for almost any pip library that isn't available in the AUR or the arch repos.

Happy jaxing!