silvesterhsu/docker-pytorch repository overview

⁠docker-pytorch

This is a Dockerfile that Jupyter Notebook automatically builds Docker images.

⁠Why build it?

This is a Dockerfile built on the pytorch/pytorch⁠ image.

Because the root is not open in the official image of jupyter, a series of operations that need to be installed with root privileges in the jupyter notebook are restricted, such as apt install.

Therefore, based on the official image of jupyter, I added additional root privileges, added support for zsh and oh-my-zsh, and commonly used plugins in jupyter notebook.

For more details,

Github: link⁠

Docker: link⁠

⁠How to run it?

Run Jupyter Notebook by default.

docker run --gpus all --restart=always --name pytorch -ti --ipc=host -p "$PORT":8888 -v "$PWD":/workspace silvesterhsu/docker-pytorch:"$TAG"

$PORT: Port mapping. It is the port that needs to link the local to the image. In docker, jupyter will open port 8888 as a web access. If the local port 8888 is not occupied, it is recommended to use 8888.

$PWD: File mapping. Project work path

$TAG: For the time being, only latest, if not filled in, the latest version is downloaded by default. The ARM version may be available in the future.

Example:

docker run --gpus all --name pytorch --restart=always -ti --ipc=host -p 8888:8888 -v ~/new_project:/workspace silvesterhsu/docker-pytorch

⁠Set password

Once you start container, an uniquetoken will be shown in the terminal.

Use the token to setup a password when you open the browser 127.0.0.1:8888.

note: The port number depends on the port you are mapping

Once the password is set and successfully logged in, jupyterLab completes the password configuration. You need to terminate and restart the lab container in the terminal.

Use control+C to stop the jupyterlab container, or start a new terminal:

docker restart pytorch

It is necessary to restart the container. After the password is stored, it needs to be restarted to apply.

Then, setting the password is complete.