Wheels for Apple Silicon M1

Announcements
1

Dear *,

I have managed to create wheels for the new Apple M1 Silicon (osx-arm64) platform. They are now available via the numba/label/wheels_experimental_m1 channel on anaconda.org:

https://anaconda.org/numba/numba/files?channel=wheels_experimental_m1

The installation instructions are as follows, in an environment that supports pip install:

pip install "numpy==1.21.*" importlib-metadata # manually install dependencies
pip install -i https://pypi.anaconda.org/numba/label/wheels_experimental_m1/simple numba

The correct llvmlite should be found and installed automatically as part of the command to install Numba.

An explicit version for NumPy is required in this case because the wheels do not (yet) pin the NumPy version and anything beyond 1.21.* isn’t/wasn’t yet supported.

If you do have capacity, please test the wheels with your library of application and report back any issues you discover here. Thank you!

2 Likes
2

Awesome news!
I set up a arm64 miniforge3 env with Apple tensorflow following the tensorflow-metal plugin instructions (sorry can’t post links)
Into this I installed numba as instructed above.
A-Ok!
Then I installed stardist via pip which has a numba requirement and previously meant having to pre-install via conda.
This went off without a hitch.
I was able to do the simple example from the StarDist readme with no issues, so looks like everything works!

3

Thank you everyone for helping to test these wheels. Since Numba 0.55.2 and llvmlite 0.38.1 are now available with wheels from PyPi, I will delete the experimental wheels from anaconda.org to avoid any confusions in the future.

2 Likes