This is a nvidia-docker image for using TensorFlow on Linux servers with JupyterHub support.
This image also contains following Python modules:
NumPy (for low-level math operations) pandas (for data manipulation) scikit-learn (for evaluation metrics) imageio (for read and write images) Matplotlib (for data visualization) Seaborn (for heatmaps)
tag | tensorflow | jupyterhub | node | python | cuda |
---|---|---|---|---|---|
0.1.0 | 1.13.1 | 0.9.4 | 10.15.3 | 3.6.7 | 10.0-cudnn7-devel |
-
Make sure you have following mountable directories
- one for all jupyterhub files
- one for all user files
-
To start a container
docker run --runtime=nvidia -d -t \
-p <port>:8000 \
-v <hub-files-dir>:/opt/jupyterhub \
-v <user-files-dir>:/home \
--name <container-name> \
xandai/tfhub-cuda:0.1.0
- To add a user
docker exec -it <container-name> bash -c "adduser <user-name>"
- To start jupyterhub
docker exec -t -d <container-name> bash -c "jupyterhub &>> jupyterhub.log"
- To enter terminal
docker exec -it <container-name> bash