Deprecated

This repo is deprecated in favor of https://github.com/diyor28/tf-aarch64. tf-aarch64 contains more wheels and more additional libraries.

tf-docker-m1

Tensorflow and Tensorflow Addons prebuilt wheels for Linux running inside Docker container on Apple Silicon

How to install

Make sure you have wheel installed

pip install wheel

Pick a tensorflow and python version from here and install using:

pip install https://github.com/diyor28/tf-docker-m1/releases/download/v1.0.0/tensorflow-2.8.0-cp39-cp39-linux_aarch64.whl

or in a Dockerfile using

RUN pip install https://github.com/diyor28/tf-docker-m1/releases/download/v1.0.0/tensorflow-2.8.0-cp39-cp39-linux_aarch64.whl

Build from source

You can also build tensorflow from source for your platform using the Dockerfile included in the repository. First clone the repository.

git clone https://github.com/diyor28/tf-docker-m1.git

Make sure to give your docker engine enough RAM since build may fail if the amount of RAM is insufficient. Also give your docker engine as many cores as possible in Docker Desktop settings to make building faster. On an 8-core (4 efficiency, 4 performance cores) M1 building takes about 5-8 hours.

Then run

cd tf-docker-m1/
docker build -t tensorflow:2.8 -f tensorflow/Dockerfile .

to build tensorflow-addons from source run

docker build -t tensorflow-addons:0.16 -f tensorflow-addons/Dockerfile .

Supported versions

Python: 3.9, 3.8
Tensorflow: 2.8
Tensorflow-addons: 0.16

NOTE: Other versions can be built and uploaded upon request.