Package Details: python-ml-dtypes 0.3.1-1

Git Clone URL: (read-only, click to copy)
Package Base: python-ml-dtypes
Description: A stand-alone implementation of several NumPy dtype extensions used in ML.
Upstream URL:
Keywords: jax tensorflow
Licenses: Apache
Groups: jax
Submitter: daskol
Maintainer: daskol
Last Packager: daskol
Votes: 0
Popularity: 0.000000
First Submitted: 2023-03-28 08:38 (UTC)
Last Updated: 2023-09-22 22:20 (UTC)

Latest Comments

Svenstaro commented on 2023-11-28 13:46 (UTC)

Moving this to [extra] as a new dep of tensorflow.

daskol commented on 2023-07-18 21:16 (UTC)

@WSDMatty Fixed. Thanks for reporting. PyBind11 is header-only library. If a project is compiled successfully and tests are passed then everything fine.

FYI This is actually a good issue and very instructive. We have already discussed package version management with @jakevdp in the context of pinnig numpy version and more extensive discussion were done in SciPy community. The issue is there is a lack of instructions in new packaging guidelines (PEP-517 and later) and packagers are usually too restrictive about dependencies. As far as I followed discussions properly, PyPA committee will clarify dependency handling in the next PEPs.

BTW I'm trying to develop an unofficial package repo with prebuilt machine/deep learning pckages like python-jaxlib or python-flash-attention.

WSDMatty commented on 2023-07-18 20:55 (UTC)

With the recent version update to pybind11 the package no longer builds.

ERROR Missing dependencies:

Editing the ml-dtypes.diff to look like

+    "numpy>=1.23.3; python_version>='3.11'",
-     "pybind11~=2.10.0",
+     "pybind11~=2.11.0",
-    "setuptools~=67.6.0",

made the package build. HOWEVER. I do not know if the newer version of pybind11 has any changes that make it incompatible. Upstream is not updated and I'm going to leave a comment for them, also.

carlosal1015 commented on 2023-05-23 00:44 (UTC)

Thanks @daskol.

daskol commented on 2023-05-08 23:41 (UTC)

Is there a way to get working with python-3.11.3?

@carlosal1015 Yes, it can be patched but I need to rebuild my repo for py311 entirely to verify the patch.

carlosal1015 commented on 2023-05-05 02:30 (UTC)

Is there a way to get working with python-3.11.3?

ERROR Missing dependencies:
    numpy~=1.23.3; python_version >= "3.11"
==> ERROR: A failure occurred in build().

daskol commented on 2023-03-29 11:23 (UTC) (edited on 2023-03-29 11:28 (UTC) by daskol)

And package built on x86_64 also doesn't work on an archlinuxarm.

And it shouldn't. You've just commented *.so. It has x86_64 platform tag.

petronny commented on 2023-03-29 11:17 (UTC) (edited on 2023-03-29 11:18 (UTC) by petronny)

Hi, this package seems not to be an any package.

It has /usr/lib/python3.10/site-packages/ml_dtypes/ which links to platform specific libraries.

$ ldd /usr/lib/python3.10/site-packages/ml_dtypes/ (0x00007ffded9b3000) => /usr/lib/ (0x00007f83c3400000) => /usr/lib/ (0x00007f83c3318000) => /usr/lib/ (0x00007f83c36e3000) => /usr/lib/ (0x00007f83c3131000)
        /usr/lib64/ (0x00007f83c377d000)

And the package built on x86_64 doesn't work on an archlinuxarm neither.

$ python
Python 3.10.9 (main, Dec 25 2022, 03:14:50) [GCC 12.1.0] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> import ml_dtypes
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/lib/python3.10/site-packages/ml_dtypes/", line 27, in <module>
    from ml_dtypes._custom_floats import bfloat16
ModuleNotFoundError: No module named 'ml_dtypes._custom_floats'

daskol commented on 2023-03-28 10:06 (UTC)

@petronny Fixed. Thanks!

petronny commented on 2023-03-28 09:53 (UTC) (edited on 2023-03-28 09:55 (UTC) by petronny)

==> Starting check()...
/startdir/PKGBUILD: line 31: pytest: command not found
==> ERROR: A failure occurred in check().

Please update checkdepends.