[ANN] Numba 0.66.0rc1 / llvmlite 0.48.0rc1

Announcements
1

Hi all,

I’m happy to announce fresh release candidates of Numba and llvmlite. Numba now upgrades to llvmlite 0.48 with LLVM 22 and restores sys.monitoring on Python 3.14.4 and later. Also, adds support for typed set containers and multi-dimensional fancy indexing. For more information about the state of the support and breaking changes for LLVM 22, please see: LLVM 22 (llvmlite 0.48+) — llvmlite 0.48.0dev0+54.ge8d38ef.dirty documentation

Numba 0.66.0rc1

Install from PyPI.org:


pip install --pre numba==0.66.0rc1

Install from anaconda.org:


conda install -c numba numba=0.66.0rc1

Release notes:

llvmlite 0.48.0rc1

Install from PyPI.org:


pip install --pre llvmlite==0.48.0rc1

Install from anaconda.org:


conda install -c numba llvmlite=0.48.0rc1

Release notes: v0.48.0rc1

Thank you again to all contributors for making this possible.

ping @RC_Testers

1 Like
2

All tests are passing for STUMPY:

1 Like
3

Hello,

I run a quick test with NuCS (GitHub - yangeorget/nucs: NuCS is a Python constraint programming library for solving Constraint Satisfaction and Optimization Problems over finite domains · GitHub) and my test suite runs three times slower!!!

Is this expected ?

Regards,

Yan Georget

4

Thanks for reporting this! I was able to trace this to regression on slice assignment. I’ve created an issue on the tracker - Performance regression on `0.66.0rc1` slice assignment · Issue #10643 · numba/numba · GitHub. This will need some more investigating for a proper fix.

5

Thank you for this in-depth investigation!

1 Like
6

Thanks, looks good on numbox, haven’t noticed any significant performance changes in tests relative to 0.65.1.

1 Like
7

[Update] A RC2 is posted for Numba

8

Hi @yangeorget ,
The 0.66.0rc2 should have fixed performance regression on slice assignment and better yet improved it than what it was before the regression. If possible, could you check the NuCS test suite with the new rc to confirm ?

9

I confirm. Thanks for the fix.

1 Like