It's pretty easy, but why not make it easier?
This is just a script that automates the installation outlined on @sypino's life changing twitter thread. Check that thread out; he also gives clear directions for installing conda, etc.
This script assumes that you have a working Apple Silicon (arm64) conda installation conda or github, or brew and are using zshell as your default shell (it has been the default for Mac since sometime in 2019).
Download the repo's zip file somewhere on your system (e.g. ~/Downloads) and unzip.
Open a terminal and cd to the folder containing the script. E.g.
> cd ~/Downloads/mac_arm_tensorflow_ez_install-main
Then run it from the terminal:
> zsh mac_arm_tensorflow_ez_install.zsh
The script will create a new environment called tflow_arm and just run through all of the steps to install pytorch in that environment. No questions or decisions. We also install jupyterlab, numpy, and matplotlib for good measure.
If all goes well you should see a bunch of text scroll by, and then see something like:
Checking installation; this may take a few.
(Taking a while is a good sign.)
Devices available:
- PhysicalDevice(name='/physical_device:CPU:0', device_type='CPU')
- PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')
Yay. Tensorflow can see the GPU! We should be good to go.Tensorflow will be available in the new tflow_arm environment. To activate, type:
conda activate tflow_arm