GPU / DirectML support not working about nixos-wsl HOT 3 CLOSED

My first recommendation is to not use conda on NixOS because it downloads pre-compiled binaries which require patching.

Also Python 3.6 is out of support in nixpkgs. Currently only Python 3.9 and 3.10 are supported.

You must add /host/usr/lib/wsl/lib to the LD_LIBRARY_PATH or the dynamic libraries fails to load

That is likely because wsl can't write the ld preload file to /etc. Maybe we can hardcode that. I found the error in the beginning of dmesg.

If you are required to use python 3.6 and conda to get your work done I would recommend to stay on Ubuntu.

PS: I personally wouldn't use conda on any Linux or Windows because the constraint resolver is slow and presents very bad recommendations. Also it usually majorly differs from Linux production environments which requires you to build two setups.

from nixos-wsl.

From https://github.com/microsoft/tensorflow-directml:

Note: Python 3.8 or newer is not currently supported. To use the official PyPi packages, the CPython interpreter is required. NumPy 1.19.4 is requires the KB4598291 to properly work on Windows.

So unless Microsoft updates tensorflow-directml to support Python >= 3.9, I don't see a way to get this to work on NixOS without major workarounds like patching conda binaries or using the python derivation from an older NixOS version

from nixos-wsl.

So unless Microsoft updates tensorflow-directml to support Python >= 3.9, I don't see a way to get this to work on NixOS without major workarounds like patching conda binaries or using the python derivation from an older NixOS version

Will close for now given that state of things. It's also my experience that the approach Microsoft took here isn't ideal since it not only does it require the specific Python packages, but also code changes. Not being able to easily collaborate with non WSL users or pull in code from GitHub without making extensive changes is unfortunate. I may be wrong or missed something here!

Thanks all!

from nixos-wsl.