Draft: Fix py-{jax{,lib},tensorflow}+cuda

changed title from Fix {--}py-jax{,lib}+cuda{--} to Fix py-jax{,lib}+cuda

marked this merge request as draft

py-tensorflow technically isn't a dependency, but jax seems to use it for profiling…

shouldn't this "dependency" also be reflected in the DAG? :/

Optional ones? It's only needed when people decide to run some "profiling" tests (that often make use of some TF functionality).

ok yes, I misunderstood your point of jax "using it" - currently, no dependency means it's not installed at all

added 9 commits

• 41a600c3...3c3fc6eb - 2 commits from branch master
• ace5a4a2 - feat(py-jax{,lib}): copy recipes from upstream@develop (2025-03-06)
• da39c44a - feat(py-jax{,lib}): enable new versions
• c9b84d87 - feat(cudnn): copy recipes from upstream@develop (2025-03-14)
• 276a6e3f - feat(cudnn): add 9.8.0
• 4377576b - feat(py-tensorflow): copy recipes from upstream@develop (2025-03-14)
• 9799f587 - fix(py-tensorflow): deps for 2.18.0
• 81fd46c2 - fix(py-tensorflow): missing conf env vars for @2.18:

Compare with previous version

added 1 commit

• 8f283565 - fix(py-tensorflow): missing conf env vars for @2.18:

Compare with previous version

added 1 commit

• e4f57455 - fix(py-tensorflow): missing conf env vars for @2.18:

Compare with previous version

added 2 commits

• d2ffbd4d - fix(py-tensorflow): missing conf env vars for @2.18:
• c4ac104f - fix(py-tensorflow): disable XNN's avxvnniint8 on gcc before 13

Compare with previous version

added 1 commit

• 349ac834 - fix(py-tensorflow): more fixes (@2.18.0)

Compare with previous version

added 1 commit

• 48ea5a88 - fix(py-tensorflow): build w/ modern cuda

Compare with previous version

(Now, this (incl. the commit on top I forgot to push below) finally worked on my very personal installation :P … we need some GPU in CI now.)

added 1 commit

• 3815db8f - fix(py-jaxlib): there might be multiple whl to install

Compare with previous version

changed title from Draft: Fix py-jax{,lib}+cuda to Draft: Fix py-{jax{,lib},tensorflow}+cuda

@earnold @pspilger FYI

@earnold please check the logs of the last run: https://gitlab.ebrains.eu/ri/tech-hub/platform/esd/ebrains-spack-builds/-/jobs/170632/artifacts/browse/spack_logs/not_installed/spack-stage-jaxsnn-10.0-a1-6hzs3m7bo7exdppegjlgjgpyzltl5wq4/

Esp. install-time-test-log.txt and summary.xml

I think both test fails are fixture test fails which are caused by changed jax RNG behavior in newer version…

→ @earnold fix or comment out (=ignore)?

added 27 commits

• 3815db8f...be1abf43 - 16 commits from branch master
• 0f85baf9 - 1 earlier commit
• 991e89ee - feat(py-jax{,lib}): enable new versions
• 711a1c97 - feat(cudnn): copy recipes from upstream@develop (2025-03-14)
• 1b54bea2 - feat(cudnn): add 9.8.0
• 93a1c3d6 - feat(py-tensorflow): copy recipes from upstream@develop (2025-03-14)
• added671 - fix(py-tensorflow): deps for 2.18.0
• 11d3abe8 - fix(py-tensorflow): missing conf env vars for @2.18:
• 71fcc9f9 - fix(py-tensorflow): more fixes (@2.18.0)
• 35ccead6 - fix(py-tensorflow): build w/ modern cuda
• 2116c57c - fix(py-jaxlib): there might be multiple whl to install
• 46516cfb - fix(jaxsnn): tests when using modern jax

Compare with previous version