Tools and scripts for friends™ who trust eachother to share a GPU box to do GPU-development on for deep learning.
Being able to run self-contained GPU development environments inside a container to avoid having libraries fight each other due to Nvidia Driver <-> Cuda Toolkit <-> cuDNN (CUDA Deep Learning Neural Network library) together with Python2( why would you?!) and Python 3, and not at least different versions of Tensorflow having different requirements... to say the least, it's a huge dependency list to maintain. Running GPU-dev inside a container where you can contain everything beside the Nvidia driver, simplifies your life and avoids screwing up your HOST OS.
You had to use a sad library which only works on a particular tensorflow version? Not a problem, run the container with correct dependencies which pulls in exactly the given version of cuDNN and Cuda Toolkit!
Since I personally prefer to run stuff as non-root, and this also doesn't screw up with file ownership when mounting my own $HOME directory from the HOST inside the container, these are helper scripts helping you to exactly do that.
A more simple route is using igorbb's gpu-dev directly, if you don't mind running everything as root inside the container.
See install.md
TODO: Fix even better usage; for now:
dz-build [image_directory=$USER/gpu-dev] (looks for folders in ~/images/ !)
dz-run <gpu_id (0|1> [port=8888] [image=$USER/gpu-dev] [cmd=/usr/local/bin/jupyter notebook --no-browser --ip=0.0.0.0]
dz-run 1 9999 # Is usually enough
dz-shell-root # gains root shell in your container run from dz-run
Current usage shown by dz-usage ↓↓↓:
If above ↑↑↑ empty, no users running containers using GPU
Remember to stop/kill your container after usage! Make sure you save notebook first!
Kill container by executing «docker stop $USER»
This help text from dz-help
- igorbb for the initial concept of getting it to run inside Docker and sharing his gpu-dev image https://hub.docker.com/r/igorbb/gpu-dev/
- See https://github.com/norrs/deeplearning-gpu-box/blob/master/CONTRIBUTORS.md