This package is not available on PyPI so that’s our first roadblock. However, Conda is only used for installing a single package, namely tensorflow-deps. Note that you do not need to use venv to follow my way, you could just as well use virtualenv (or no virtual environment at all if you are such inclined, though I would not recommend that).Īpple’s instructions for installing TensorFlow will tell you to install Miniforge which is a minimal installer for Conda.

This post explains how I managed to do that. This setup works well for me, and I’d like to keep it. For managing virtual environments, I use venv which is Python’s relatively new, built-in implementation of virtual environments. But, for all my projects, both work and personal, I use Homebrew for managing non-Python dependencies and pip for Python-dependencies. In this short post, I will show you how to get TensorFlow up and running with GPU support on your Apple Silicon Mac without installing Miniforge or anything else related to Conda!įirst a quick bit of background. If you are like me and already have a setup for running virtual environments that does not involve Conda, and you’d like to keep it that way, Apple’s instructions are not very helpful. Unfortunately, Apple’s installation instructions are not very clear, and they expect you to use a mix of conda and pip.

With the release of Apple Silicon Macs, we finally have a way to (easily) install and run TensorFlow with GPU support on macOS. TLDR Run brew install hdf5, then pip install tensorflow-macos and finally pip install tensorflow-metal. TensorFlow with GPU support on Apple Silicon Mac with Homebrew and without Conda / Miniforge

Metal gpu support