arm64-docker-pytorch-tensorflow

Dockerfile of a Deep Learning development environment with PyTorch and Tensorflow for arm64 architecture, especially Applie Silicon (M1+) Macs.

• Docker image: https://hub.docker.com/r/sonoisa/deep-learning-coding
  • docker pull sonoisa/deep-learning-coding:pytorch1.6.0_tensorflow2.3.0
  • docker pull sonoisa/deep-learning-coding:pytorch1.12.0_tensorflow2.9.1

This Dockerfile is a merged version of the Dockerfile for PyTorch and the Dockerfile for TensorFlow described in the following site. https://community.arm.com/developer/tools-software/tools/b/tools-software-ides-blog/posts/aarch64-docker-images-for-pytorch-and-tensorflow

How to build

To build on 16GB M1 MacBook Pro

• Make sure to allocate 12GB of memory and 4 cpu to Docker.
• The build may fail due to lack of memory, so restart the Docker service itself and run the build immediately after that.

$ ./build.sh --build-type full --jobs 2

To build on 64GB M1 Max MacBook Pro

• Make sure to allocate 48GB of memory and 8 cpu to Docker.

$ ./build.sh --build-type full --jobs 8 --bazel_memory_limit 24576