Installing_Tensorflow_on_Mac_M1

https://medium.com/dive-into-ml-ai/installing-tensorflow-natively-on-mac-m1-in-2022-1357e9b7a201

Installing tensorflow natively on mac-M1 in 2022
As things stand on May 2022

Since the original release by Apple in November 2020 of first Macs with an Arm-based M1 chip, there has been a constant struggle to install tensorflow natively on these machines.

A single internet search and one can come across pages and pages full of people complaining about this. Numerous workarounds have been proposed over the months, none without their own complications. A glimpse can be seen on the discussion on Apple’s developer forum.

I bought my M1-machine back in May, 2021 and spent a lot of time struggling, before giving up on it myself as well. A year on and a few minor hiccups later, I could finally got it working. Here are the involved steps.

System Configuration
macOS Monterey 12.3.1
pyenv local 3.9.1
Procedure

Simply trying pip install tensorflow-macos, resulted in following installation issue with h5py
error: Unable to load dependency HDF5, make sure HDF5 is installed properly

It was resolved following the below solution from link.

brew install hdf5
export HDF5_DIR="$(brew --prefix hdf5)"
pip install --no-binary=h5py h5py 
Moving on, installing tensorflow-io-gcs-filesystem threw another error,
Unable to find installation candidates for tensorflow-io-gcs-filesystem (0.25.0)

To resolve this I had to manually install tensorflow-io-gcs-filesystem from the wheel file as proposed here.

git clone https://github.com/tensorflow/io.git
cd io
python setup.py -q bdist_wheel
python -m pip install --no-deps dist/<wheel-file-from-last-step>
Of course, this is still far from ideal but I find it much better than the earlier solutions requiring workarounds with miniforge, etc.

Disclaimer: The installation was done using pyenv with local python 3.9.1. There might still be issues with earlier python versions. 
In case there are issues working with pyenv on macOS please check out my article here.