Numba 0.59.0 and llvmlite 0.42.0 releases are published. These new releases add support for Python 3.12. For details about new features and bug fixes, please see the release notes at:
To install using
$ conda install -c numba numba=0.59.0 llvmlite=0.42.0
To install using
$ pip install numba==0.59.0 llvmlite==0.42.0
I am looking to learn from other people’s recent overall experience, so don’t get me wrong.
For me, numba is currently defunct due to it’s official website examples not working as advertized on my i7-8700 setup with numba 0.58 and python 3.10 on linux (the speedups are only negligibly x2 on most official examples), and in particular after the following personal experience at Elapsed running time of parallel jit for the logistic regression documentation example is dozens of seconds · Issue #9395 · numba/numba · GitHub. I wonder if this is the same overall friction as what other people are experiencing, obviously, it can be just bad luck with library versions or glitched builds of them.
I switched to writing cython and letting numpy run its natural course of threaded execution on some of its operations, but happy to learn from other people’s experience in case I come back later.
Since these days numpy or the underlying BLAS libraries use threading these days for some operations, I wonder why is numba kind of not merging into the numpy team, or merging release cycles with them, to streamline the stability and performance of numeric computation in the python world while is seamingly very consistently touting its numpy integration as per the numba website.
Just in case you have some helpful comments on the overall (my Github documented issue aside, as this is not a message trying to prompt more attention for that particular experience or issue).