Hi folks 👋
This is simply a setup instruction for machine learning required packages, Python and TensorFlow on Apple Metal M1.
The latest Mac ARM M1-based machines have considerably better machine learning support than their previous Intel-based counterparts and yet it is exciting to try some casual ML models using the neural engine in this chip. While most of the CUDA packages architecture is different than the M1, installing ML packages could be a pain at the bottom!
This instruction aims to simply and craft a step-by-step guide to setup using MiniForge.
Note: PyTorch is currently struggling to catch up with M1 and there is no straight solution to using accelerated GPU with Python, feel free to pull this repo and add any other update/solution you might have.
Homebrew is a package manager for the Mac, kind of similar to yum or apt-get for Linux. Copy and paste below line in a macOS Terminal to get things rolling.
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"
🚨 follow the steps when it finished, requires to echo command.
sudo chown -R $(whoami) /opt/homebrew /opt/homebrew/share/zsh /opt/homebrew/share/zsh/site-functions /opt/homebrew/var/homebrew/locks
then:
chmod u+w /opt/homebrew /opt/homebrew/share/zsh /opt/homebrew/share/zsh/site-functions /opt/homebrew/var/homebrew/locks
Now, you need to install the xcode-select command-line utilities. Use the following command to install:
xcode-select --install
🚨 If this command didn't work, you should install XCode from the App Store.
It's time to put Homebrew to work, let's install MiniForge using Homebrew with this line:
brew install miniforgeMiniForge is based on Anaconda but it emphasize on different CPU architectures such as M1 which is our current concern.
Jupyter notebooks are always handy and good to have, so along with your IDE of your choice, install using the below command:
conda install -y jupyter
I found at this file authored by Jeff Heaton very useful. It installs some of very handy libraries at once, such as Pandas, SciPy, Matplotlib, etc.
I highly recommend installing them using the .yml file in this repo and run the command below which creates an environment called tensorflow and install all the dependencies in the file:
👉 I do highly recommend installing MiniConda along with MiniForge before proceeding as it helps keeping your PATHs organized.
conda env create -f tensorflow-apple-metal.yml -n tensorflow
Once the setup is done, access to this environment using the following command:
conda activate tensorflow
🚨 If the above command shows error, try conda init {SHELL} such as zsh or bash and run the activate again. If that didn't work too, then try conda info --envs, copy the path of tensorflow env and use conda activate /PATH.
Let's also add Jupyter to this environment:
conda install nb_conda
Alright it's time to register our newly created environement to the kernel using this command:
python -m ipykernel install --user --name tensorflow --display-name "Python 3.9 (tensorflow)"
🚨 Make sure your environment is activated and you are in (tensorflow).
Almost there! Let's just make sure everything is setup and works fine! Jump in to a new Jupyter notebook using this command:
jupyter notebook
And now paste and run this code to validate your setup:
import sys
import tensorflow.keras
import pandas as pd
import sklearn as sk
import tensorflow as tf
print(f"TensorFlow Version: {tf.__version__}")
print(f"Keras Version: {tensorflow.keras.__version__}")
print()
print(f"Python {sys.version}")
print(f"Pandas {pd.__version__}")
print(f"Scikit-Learn {sk.__version__}")
gpu = len(tf.config.list_physical_devices('GPU'))>0
print("GPU is", "available" if gpu else "NOT AVAILABLE")Wa-la! Here is your new glory M1 running latest versions of Machine Learning packages and waiting for you to create awesome stuff.