Ubuntu 14.04 + Torch + CUDA (optional)
Can be used to:
- Run an iTorch notebook server
- Execute Torch scripts
In order to use this image you must have Docker Engine installed. Instructions for setting up Docker Engine are available on the Docker website.
This image can be built on top of multiple different base images derived from Ubuntu 14.04. Which base you choose depends on whether you have an NVIDIA graphics card which supports CUDA and you want to use GPU acceleration or not.
If you are running Ubuntu, you can install proprietary NVIDIA drivers from the PPA and CUDA from the NVIDIA website. These are only required if you want to use GPU acceleration
If you are running Windows or OSX then you will find it difficult/impossible to use GPU acceleration due to the fact that containers run in a virtual machine. Use the image without CUDA on those platforms.
Firstly ensure that you have a supported NVIDIA graphics card with the appropriate drivers and CUDA libraries installed.
Build the image using the following command:
docker build -t torch cuda-7.5You will also need to install nvidia-docker, which we will use to start the
container with GPU access. This can be found at
NVIDIA/nvidia-docker.
Build the image using the following command:
docker build -t torch no-cudaNV_GPU=0 nvidia-docker run --rm -it --volume=/path/to/notebook:/root/notebook \
--env=JUPYTER_PASSWORD=my_password --publish=8888:8888 torchIf not using the CUDA base, this command should be modified slightly to use
the docker command directly instead of the wrapper script:
docker run ...Replace /path/to/notebook with a directory on the host machine that you would
like to store your work in.
Point your web browser to localhost:8888 to start using the iTorch notebook.
It is also possible to run Torch programs inside a container using the th
command. For example, if you are within a directory containing some Torch
project with entrypoint my-script.lua, you could run it with the following
command:
NV_GPU=0 nvidia-docker run --rm -it --volume=$PWD:/root/notebook torch th my-script.luaThe important thing to remember is that the default working directory in the
container is /root/notebook. By creating a volume which maps the current
working directory to that location in the image, th is able to find and
run our script.
If you are running on a Linux host, you can get code running inside the Docker container to display graphics using the host X server (this allows you to use OpenCV's imshow, for example). Here we describe a quick-and-dirty (but INSECURE) way of doing this. For a more comprehensive guide on GUIs and Docker check out http://wiki.ros.org/docker/Tutorials/GUI.
On the host run:
sudo xhost +local:rootYou can revoke these access permissions later with sudo xhost -local:root.
Now when you run a container make sure you add the options -e "DISPLAY" and
--volume="/tmp/.X11-unix:/tmp/.X11-unix:rw". This will provide the container
with your X11 socket for communication and your display ID. Here's an
example:
NV_GPU=0 nvidia-docker run --rm -it \
-e "DISPLAY" --volume="/tmp/.X11-unix:/tmp/.X11-unix:rw" \
--volume="$PWD:/root/notebook" torch \
th -e 'cv = require"cv";require"cv.highgui";cv.namedWindow{"Foo"};cv.waitKey{0}'You can create a notebook.json config file to customise the editor. Some of the
options you can change are documented at
https://codemirror.net/doc/manual.html#config.
Let's say that you create the following file at /path/to/notebook.json:
{
"CodeCell": {
"cm_config": {
"lineNumbers": false,
"indentUnit": 2,
"tabSize": 2,
"indentWithTabs": false,
"smartIndent": true
}
}
}Then, when running the container, pass the following option to mount the configuration file into the container:
--volume=/path/to/notebook.json:/root/.jupyter/nbconfig/notebook.jsonYou should now notice that your notebooks are configured accordingly.