/ansible-ml-env

What it does

It installs:

  • Tensorflow+docker
  • PyTorch
  • Darknet
  • Python environments for testing things out

How to run

$ ansible-galaxy install -r playbook-requirements.yml
$ ansible-playbook -i host, playbook.yml

How to update these roles

Subscribe to RSS feed of tensorflow releases: https://github.com/tensorflow/tensorflow/releases.atom
Update version numbers given here: https://www.tensorflow.org/install/gpu#install_cuda_with_apt

PyTorch releases RSS: https://github.com/pytorch/pytorch/releases.atom

To-do list

TODO: multiple cuda installations:
https://devtalk.nvidia.com/default/topic/493290/cuda-programming-and-performance/multiple-cuda-versions-can-they-coexist-/post/3532363/#3532363
https://stackoverflow.com/questions/41330798/install-multiple-versions-of-cuda-and-cudnn/53345192#53345192
https://docs.nvidia.com/deploy/cuda-compatibility/index.html

https://stackoverflow.com/questions/50622525/which-tensorflow-and-cuda-version-combinations-are-compatible

TODO: automate testing they all see the GPU

~/.python_venv/tensorflow2-beta/bin/python -c 'import tensorflow as tf; print(tf.test.gpu_device_name())'
~/.python_venv/pytorch-py37/bin/python -c 'import torch; print(torch.cuda.is_available())'

TODO: -i inventory.yml, https://docs.ansible.com/ansible/latest/user_guide/intro_inventory.html

all:
  hosts:
    mail.example.com:
  vars:
    group_all_var: value