...This is a work in progress!
A collection of Dockerfiles for building docker containers used to spawn single-user notebooks with DockerSpawner in JupyterHub. All notebooks are GPU enabled (CUDA 8.0, cuDNN v5.1), and there are different containers supporting several popular deep-learning frameworks (e.g., Tensorflow, Torch, Caffe, etc.).
The base notebook was created by modifying nvidia-cuda-devel to work as a single-user notebook with jupyterhub, and then adding OpenCV.
https://gitlab.com/nvidia/cuda/blob/ubuntu14.04/8.0/runtime/cudnn5/Dockerfile https://gitlab.com/nvidia/cuda/blob/ubuntu14.04/8.0/devel/Dockerfile
Most of the framework notebooks (e.g., Tensorflow, Torch, etc.) were created by copying https://github.com/floydhub/dl-docker, and then making minor modifications for compatibility with the base-notebook.
- Ubuntu 14.04
- CUDA 8.0 (GPU version only)
- cuDNN v5.1 (GPU version only)
- iPython/Jupyter Notebook (including python2 and python3 kernals)
- Numpy, SciPy, Pandas, Scikit Learn, Matplotlib
- OpenCV
https://developer.nvidia.com/deep-learning-frameworks
- Pytorch
- Tensorflow, with the option to use Keras or TFLearn
- Torch (includes nn, cutorch, cunn and cuDNN bindings), and iPython/Jupyter Notebook with the itorch kernal.
- CNTK v2.0-GPU-1bit-SGD
- NVCaffe (NVIDIA's fork of Caffe)
- Caffe
[ ] - Caffe2
[ ] - Theano which can be used with or without Keras or Lasagne
[ ] - Chainer
[ ] - MXNET
[ ] - Digits