vgsml / gpu-jupyterhub Goto Github PK

This repo contains the docker-compose.yml for defining the dependencies to run the Jupyter Notebook Images described in the Dockerfiles under Jupyter_Images. There is one Jupyter Notebook image, CUDA10.0 and Ubuntu18.04 based. It will be spwaned from a Jupyterhub which is defined in the Dockerfile.jupyterhub under the folder Jupyterhub_Image. For more details visit our website https://www.dlm.med.fau.de/setting-jupyterhub-deep-learning/

Edited the repo:

• removed the cuda 9 image
• restructured the docker-compose.yml --> you will need a folder called secets containing the env_files mentioned in the yml-env_file --> will need a folder under Jupyterhub_Image called ssl containing your ssl certs for building and running the Jupyterhub.

The Jupyter_Image (jupyternotebook) has new features and some removed features: Python2 is gone R is available Python3 is available SOS is available

The workhorse for deep-learning is Python3 based. It contains Tensorflow-GPU 1.14, Keras 2.2.4 Pytorch 1.2.0 and Fastai 1.0.57

-Jupyterhub Image definition with DockerSpawner for spawning Jupyternotebooks. Localauthentication is removed and replaced by OAuth with Github

• Data is persisted: --> -locally (Docker-Volume --> cookie secrets) --> host machine and pesronal data is mapped into the container for pre spwan hook (see into jupyterhub_config.py)

-Spawned Images run in single Docker-Containers
- Data is persisted: --> user based: -locally on nvme/ssd (Docker-Volume bind mount--> host_path:/home/Deep_Learner/private/local) and per network associated folder
(host_path:/home/Deep_Learner/private/network). --> commonly shared: - per network associated folder (net_share:/home/Deep_Learner/shared).

Added ftp server mapping to an uploads folder in host_path/uploads:/home/Deep_Learner/shared/uploads