Docker image supporting fastai, pytorch, tensorflow, jupyter, and anaconda.
The main purpose of this repository is to be able to pull an run this image from a variety of environments, including AWS, GCP, and my home computer.
This image is published to DockerHub as dennisobrien/deeplearning-fastai.
It is assumed that it is running on a host that has a few things already installed and enabled.
- nvidia drivers
- CUDA 9.0, 9.2, 10.0
- nvidia-docker
First set the current directory to the location of your notebooks you want to serve. Then start the docker container.
$ docker run -it --publish 8888:8888 --publish 6006:6006 \ -v ${PWD}:/home/jovyan/workspace \ -v ${HOME}/.fastai:/home/jovyan/.fastai \ dennisobrien/deeplearning-fastai:latest \ start-notebook.sh --notebook-dir=/home/jovyan/workspace
A little explanation of these parameters.
- We are opening a few ports.
- 8888 for Jupyter
- 6006 for TensorBoard
- We are mounting a few volumes.
~/workspacein the container will be mapped to the current directory in the host.~/.fastaiin the container will be mapped to~/.fastaiin the host.
- We instruct Jupyter to use
~/workspaceas the root of the notebook directory.
This image is built by DockerHub so it is not necessary to build it locally. But if you need to for whatever reason, use this command:
$ docker build -t dennisobrien/deeplearning-fastai:latest .
This step is not necessary since the image is built automatically after a push to the GitHub repository.
$ docker push dennisobrien/deeplearning-fastai:latest