Allows you to use jupyterlab remotely with RTX 6000 without complex complex configuration:)

Author: Kanta Yamaoka, Undergrad student of Osaka Prefecture University

This is for:

• machine with Nvidia RTX A6000
  • Using image nvidia/cuda:11.7.1-base-ubuntu20.04
  • Please make sure the cuda version is compatible with your gpu node
• docker and docker-compose are configured

Getting started

SSH into your gpu machine, and ./dc-start.sh. This provides port number for the jupyterlab.

Given that port number, open new terminal on your CLIENT, and use the following:

ssh -L 8888:127.0.0.1:{port_provided_by_above} {your_user_name}@{your_gpu_node_with_a6000}

Now, open your browser and go to:

http://localhost:8888/

Voila! Now you can access to gpu with jupyterlab from your own computer! Enjoy RTX A6000!

Note

• I haven't added mapping from the local storage on your gpu node to docker container
  • I will add it soon!
• You can also modify docker-compose.template.yaml directly.
• It chooses port on your gpu machine automatically for jupyterlab.
• You can choose devices at dc-start.sh; please have a look at the script.

Overview

• dc-start.sh starts and provides port
• dc-stop.sh stops docker container properly

├── README.md
├── dc-start.sh
├── dc-stop.sh
├── docker-compose.template.yaml
├── dockerfiles
│   └── min-pytorch
│       ├── Dockerfile
│       ├── req_ml_backend.txt
│       └── req_ml_tools.txt

Acknowledgements

Many thanks to websites on the internet that helped making this tool! (They are written in the source code.)