[Issue]: failure to handle missing python module joblib

[trixirt]

Problem Description

On RHEL and SUSE, joblib is not available in the distro. This import will fail https://github.com/ROCm/hipBLASLt/blob/develop/tensilelite/Tensile/Common/Parallel.py#L31

For rocBLAS, this is handled here https://github.com/ROCm/Tensile/blob/develop/Tensile/Utilities/ConditionalImports.py#L30

Operating System

RHEL, SUSE

CPU

ALL

GPU

AMD Instinct MI250

Other

No response

ROCm Version

ROCm 6.2.3

ROCm Component

No response

Steps to Reproduce

No response

(Optional for Linux users) Output of /opt/rocm/bin/rocminfo --support

No response

Additional Information

No response

[ppanchad-amd]

Hi @trixirt. Internal ticket has been created to fix this issue. Thanks!

[davidd-amd]

@trixirt I see this in the docs:

Joblib has no mandatory dependencies besides Python (supported versions are 3.8+

I suspect there is an older version of python on RHEL8 presumably and SUSE. Can you confirm distro versions and the python version for those distros. We will patch this in hipblaslt. Sadly with a single process it will take an incredibly long time to build.

[trixirt]

Yes, it takes a long time to build.

Here is the workaround for doing a single threaded. https://src.fedoraproject.org/rpms/hipblaslt/blob/rawhide/f/0001-hipblaslt-handle-missing-joblib.patch This is for EPEL 10 here is a preview https://copr.fedorainfracloud.org/coprs/g/rocm-packagers-sig/cs10-pytorch-preview/package/hipblaslt/ There is no plan for ROCm in-box in RHEL/EPEL 8 or 9.

Since I filed this issue, SUSE has made joblib available on all of the versions we build rocm for.

[assistant-librarian[bot]]

This issue has been migrated to: https://github.com/ROCm/rocm-libraries/issues/288

[idass1990]

Closing the issue in this repo. Please refer to the migrated issue for updates.