Scaling multi-node multi-GPU workloads
The commands below assume Python is Python 2, so you may need to change the python invoked in the script if your default Python is Python 3
# Assuming we are starting in repository root
ROOT="$PWD"
cd /tmp
git clone https://github.com/SaMnCo/dl-training-datasets
cd imagenet/ilsvrc14
./00-build.sh "$ROOT/data"
TODO: make own script for this, without the need for the user to specify Python 2.
[https://github.com/pytorch/examples/tree/master/imagenet](Follow steps here).
TODO: copy in relevant files
TODO: make user-independent aws-magic analogue, pull in docker for pytorch.