dl-project

A fairly opinionated template for starting new deep-learning projects quickly and iterating efficiently

Getting Started

TODO: Add proper setup instructions

• Clone this repo

• chmod +x run.sh

• chmod +x deploy.sh

• Set up the VM:

  • https://cloud.google.com/compute/docs/gpus/add-gpus#install-gpu-driver
  • make requisite dirs wherever u want
  • docker-machine create --driver google --google-project tai-dcn --google-zone us-west1-b --google-use-existing training-vm
  • Add firewall exceptions as appropriate (5000 and 8888)
  • Set default runtime on server to be nvidia runtime
  • Set up mutagen if desired?
    • https://stackoverflow.com/questions/32189001/how-to-replace-the-default-ssh-key-to-get-access-to-docker-machine (user is docker-user)
    • https://mutagen.io

• source the remote env

• build the image and deploy it

• Actually, I thnk get rid of Mutagen, it's more trouble than its worth

• Replace with: https://gist.github.com/mjul/bd0fbffe2f94bbf5a070d8d555f3414a ??