Cannot build via pip on macOS ARM64 M1 Apple Silicon

[KTRosenberg]

I am running macOS 11, python 3.9, and am using ARM64 hardware (Apple Silicon M1). It's a MacBook Air.

Trying to install tensor flow via pip
pip3.9 install tensorflow-macos==2.6.0

results in the following error:

ERROR: Could not find a version that satisfies the requirement h5py~=3.1.0 (from tensorflow-macos) (from versions: 2.2.1, 2.3.0b1, 2.3.0, 2.3.1, 2.4.0b1, 2.4.0, 2.5.0, 2.6.0, 2.7.0rc2, 2.7.0, 2.7.1, 2.8.0rc1, 2.8.0, 2.9.0rc1, 2.9.0, 2.10.0, 3.0.0rc1, 3.0.0, 3.1.0, 3.2.0, 3.2.1, 3.3.0, 3.4.0, 3.5.0, 3.6.0)
ERROR: No matching distribution found for h5py~=3.1.0

h5py doesn't currently have a wheel available. I can install from source, for example, by installing using home-brew and the following command:
HDF5_DIR=/opt/homebrew/Cellar/hdf5/1.12.1 pip3.9 install --no-binary=h5py h5py==3.1.0
This just installs without the binary.

However, there's no way for me to specify this in the tensor flowinstallation via pip, so it won't work.

How can we resolve this issue?

[mohantym]

Hi @KTRosenberg ! Could you refer these thread for installation. I suggest for 2.5 branch as 2.6 branch is having some issues during model creation in MacOS. You can also build from source referring this link. Thanks!

[KTRosenberg]

Oh I see. So does this absolutely require conda? I'm in a challenging situation in which a few libraries I depend on entirely have to install tensorflow, so it's not simple to migrate to using conda instead of basic pip, as some of those libraries don't behave under conda.
In other words, the problem isn't just tensorflow. I'm sure the solutions you linked are great standalone. The library interop is not likely to work though for something like this.

For what it's worth, I am using Spacy, which is seemingly having its own trouble with getting transformers-based models and installations working, and I am also attempting to use a library called "coreferee," which seems much less aware of the Apple Silicon changes required. It's a little bit of a mess.

Do you anticipate the basic pip installation being usable again sometime in the near future? I imagine things are still transitioning over.
If that's the case, I could hang on a bit longer on an intel-based machine. Hopefully fixes are on the order of a couple of months max? Do you have a sense of how things are coming along?

[mohantym]

@KTRosenberg !It is always advised to test the ongoing work in virtual environments to avoid library conflicts . But you can create virtual environment without Conda too. Attaching relevant thread for reference .Thanks!

[KTRosenberg]

Yes, I understand.

Are the tensorflow people going to work on a simpler installation process using just pip eventually? Not everyone likes conda.

[mohantym]

@KTRosenberg ! I feel positive about that too. Feel free to close this issue if it helped. Thanks!

[google-ml-butler]

Are you satisfied with the resolution of your issue?
Yes
No