Arch Linux User Repository

@rekman for the jax problem just use export XLA_FLAGS=--xla_gpu_cuda_data_dir=/opt/cuda to fix it (you can add it to .bashrc file for example). But, I got other problems in the test process

I think this will have issues rebuilding for Python 3.13 due to issues with building the latest python-jax. As I understand it those are due to upstream packaging and will have to wait for a new upstream release.

@hashworks @toynbeeidea The issue is related to other package python-dm-tree which is testing dependency for python-chex. Solution is to rebuild python-dm-treee (which depends on shared objects).

Installed broken since 10/8, see comments for fix

This package is not broken. Rebuilt python-dm-tree dependency.

Check now fails due to abseil-cpp update: ImportError: libabsl_strings.so.2401.0.0: cannot open shared object file: No such file or directory.

It should use /usr/lib/libabsl_strings.so.2407.0.0.

@daskol There was the release https://github.com/jax-ml/ml_dtypes/releases/tag/v0.4.0 built against numpy==2.0.0rc1, so I'm not entirely sure what's going on here. More investigation required.

@sbrl The issues occurs in check stage. You can just skip it with --nocheck option.

It seems that python-jax or python-jaxlib cause that exception. They depend on python-ml-dtypes. And all their binary extensions depend on version python-numpy which breaks ABI in version 2. Are these packages built with python-numpy>=2?

@timostolz It seems that you installed python-dm-tree with Python 3.11 then updated to Python 3.12. I've already bumped python-dm-tree version to trigger rebuilt with Python 3.12 in June.

The build was working after explicitely re-building and re-installing aur/python-dm-tree

Since the lastest update, builds fail for me.

===================================================================================== ERRORS ======================================================================================
__________________________________________________________________ ERROR collecting chex/_src/dataclass_test.py ___________________________________________________________________
ImportError while importing test module '/home/timo/.cache/yay/python-chex/src/chex-0.1.86/chex/_src/dataclass_test.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
/usr/lib/python3.12/importlib/__init__.py:90: in import_module
    return _bootstrap._gcd_import(name[level:], package, level)
chex/_src/dataclass_test.py:33: in <module>
    import tree
E   ModuleNotFoundError: No module named 'tree'
============================================================================= short test summary info =============================================================================
ERROR chex/_src/dataclass_test.py
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Interrupted: 1 error during collection !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
================================================================================ 1 error in 1.13s =================================================================================
==> FEHLER: Ein Fehler geschah in check().
    Breche ab...
 -> Fehler beim Erstellen: python-chex-exit status 4

(aur/python-dm-tree is installed)

This package doesn't build for me. I get the following error:

A module that was compiled using NumPy 1.x cannot be run in
NumPy 2.0.0 as it may crash. To support both 1.x and 2.x
versions of NumPy, modules must be compiled with NumPy 2.0.
Some module may need to rebuild instead e.g. with 'pybind11>=2.12'.

If you are a user of the module, the easiest solution will be to
downgrade to 'numpy<2' or try to upgrade the affected module.
We expect that some modules will need time to support NumPy 2.


Traceback (most recent call last):  File "/usr/bin/pytest", line 8, in <module>
    sys.exit(console_main())
  File "/usr/lib/python3.12/site-packages/_pytest/config/__init__.py", line 206, in console_main
    code = main()
[truncated for brevity - see full transcript below]
  File "/home/sbrl/.local/lib/python3.12/site-packages/ml_dtypes/__init__.py", line 32, in <module>
    from ml_dtypes._finfo import finfo
  File "/home/sbrl/.local/lib/python3.12/site-packages/ml_dtypes/_finfo.py", line 19, in <module>
    from ml_dtypes._ml_dtypes_ext import bfloat16
Traceback (most recent call last):
  File "/usr/lib/python3.12/site-packages/numpy/core/_multiarray_umath.py", line 44, in __getattr__
    raise ImportError(msg)

Full transcript: https://pastebin.com/raw/TGLcVCZF

I tried rebuilding python-dm-tree [from the aur] as @daskol suggested, but this has not helped my issue.

Obviously there's an issue here with Numpy and ml-dtypes compiling against it, but I am kinda stumped as to how to resolve this.

I have python-ml-dtypes@0.4.0-2 installed via pacman. The same package appears to be version 0.4.0 on PyPi, so this suggests to me that either this package is linking against ml-dtypes when it shouldn't, or ml-dtypes has yet to be updated to support Numpy 2.0, which was released on 16th June 2024.

uname -a Linux DEVICE_NAME 6.9.7-artix1-1 #1 SMP PREEMPT_DYNAMIC Fri, 28 Jun 2024 18:11:28 +0000 x86_64 GNU/Linux, python --version Python 3.12.4

My original intention was to install python-transformers.