How To Install TensorFlow on M1 Mac

Abstraction: Step-by-step TensorFlow setup on Apple M1 ARM64 chip

Key points:

• Apple released the ARM64 M1 chip in November 2020; as of writing, Apple had released TensorFlow 2.5 for Apple Silicon
• Install chain: Xcode CLI tools → Miniforge (arm64) → conda virtual env (Python 3.8) → tensorflow-deps, tensorflow-macos, tensorflow-metal plugins
• The tensorflow-metal package enables GPU acceleration via Apple's Metal API on M1
• Benchmark on MNIST with a 2-layer CNN, batch size 128, 12 epochs completed in ~1 min 57 s on M1
• Miniforge is required (not Anaconda) because it provides natively compiled ARM64 packages

Connections: Tensorflow · Apple · Deep Learning · Hardware Acceleration

Source: https://caffeinedev.medium.com/how-to-install-tensorflow-on-m1-mac-8e9b91d93706