Carmen's avatar

Tensorflow on a M3 macOS

As of early 2024, TensorFlow has native support for Apple Silicon (M1/M2/M3 chips). This means you can run TensorFlow efficiently on your M3 Mac without needing to use Rosetta 2 for translation.

  • Installation is simpler now. Apple has worked with the TensorFlow team to create a package called tensorflow-macos, which is optimized for Mac systems with Apple Silicon.

  • Metal support. TensorFlow on Mac can use Apple's Metal framework for GPU acceleration, which is optimized for Apple Silicon.

Before installation

  • Make sure you have Xcode Command Line tools (Xcode CLT) installed by typing xcode-select -p in your terminal.

More info Install Xcode Command Line tool

  • Make sure you have a pyenv or conda for environment management.

I have followed Conda's tip: "If you are just starting out, we recommend installing Conda via the Miniconda installer." Choose based on your experience..

I just finished an AI/ML bootcamp so I'm still a 🥖 bun in the oven 😅.

Download the latest Miniconda installer for macOS with Apple Silicon. After the installation is finished, open your terminal and type the command conda. You will see (base) in your terminal, which is the default environment.

🔔 Tip

It's a good idea to have jupyter installed there.
(base) ➜ ~ conda install -y jupyter

Installation

Option 1: Install packages one by one on a new environment

1. Create and activate a virtual environment: conda create -n ENV_NAME

2. Upgrade pip: python3 -m pip install --upgrade pip

3. Install TensorFlow packages: python3 -m pip install tensorflow-macos tensorflow-metal

For example, you can create an environment with a python's version

conda create -n ENV_NAME python=<version> then activate it

such as:

conda create -n tf python=3.11`
....
conda activate tf

(tf) python3 -m pip install tensorflow-macos
(tf) python3 -m pip install tensorflow-metal

Option 2: Bundling packages on a yml file.

Put together a yml file similar to tensorflow-m3-macos-env.yml

Note: You do not need to install a specific package called tensorflow-deps for running TensorFlow on an M3 Mac as it used to be months ago. In my case, I decided to bundle the packages in a yml file, where I added jupyter notebook and some data science / machine learning packages I use often (pandas numpy matplotlib scikit-learn). Yet, trying to install tensorflow with python 3.12didn't work, so I downgraded to python 3.11

conda env create --name tf -f tensorflow-m3-macos-env.yml

Create environment

conda create -n ENV_NAME

For example, you can create an environment with a python's version

conda create -n ENV_NAME python=<version>

such as: conda create -n tf python=3.11, and then to activate run

conda activate tf

You will see (tf)in your terminal, indicating that you are currently on that environment.

You can test tensorflow quickly by importing it on a python environment in your terminal.

(tf) ➜  tensorflow-on-m3-macos git:(main) ✗ python3
Python 3.11.9 | packaged by conda-forge | (main, Apr 19 2024, 18:34:54) [Clang 16.0.6 ] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import tensorflow as tf
>>> print(tf.__version__)
2.16.2
>>>