I have prepared a simple Ansible script which will enable you to convert a clean Ubuntu 18.04 image (as supplied by Google Compute Engine or PaperSpace) into a CUDA 10, PyTorch 1.0 preview, fastai 1.0.x, miniconda3 powerhouse, ready to live the (mixed-precision!) deep learning dream.

I built this script specifically in order to be able to do mixed-precision neural network training on NVIDIA’s TensorCores. It currently makes use of the vxlabs.com build of PyTorch 1.0, because we need full CUDA 10 for the new TensorCores.

When I run this in order to configure a V100-equipped paperspace machine with 8 cores and 30GB of RAM, it takes about 13 minutes from start to finish.

Here’s a 20x sped up video showing the script doing it’s work on a GCE V100 machine, also with 8 cores and 30 GB RAM:

After running the script, you’ll be able to ssh or mosh in, type conda activate pt, and then start your NVIDIA-powered deep learning engines.

You can find the whole setup, including detailed instructions, at the ansible-ubu-to-pytorch github repo.

Updates

2018-11-24

Updated to latest 2018-11-24 build of PyTorch 1.0 preview with the new magma 2.4.0 packages.

To update an existing install, you can either just re-run the whole playbook, or you can run just the miniconda3-related tasks like this:

1
ansible-playbook -i inventory.cfg deploy.yml --tags "miniconda3"