Ersa (Jupyter Remote GPU)

This project provides an ipython magic extension, to enable training pytorch models on remote GPU servers.

This project is named ersa after one of the moons of Jupiter.1

Getting Started

# clone the repo
git clone git@github.com:osmhpi/ersa.git
cd ersa
# create a new virtual environment
python3 -m venv venv
source ./venv/bin/activate
# install dependencies and additional dependencies for the demo
pip3 install .[demo]

Forward port 6001 to a remote server with a running server instance.

ssh -L 6001:127.0.0.1:6001 my-gpu-server.com

You should now be able to run the client.ipynb notebook.

The first run will probably be quite slow, because the server will recreate your local venv, including compiling the correct python version.

Server Setup

You need pyenv installed. make sure to have the following packages installed

• libsqlite3-dev
• libsqlite3-dev liblzma-dev libctypes-ocaml libreadline-dev libbz2-dev

libffi-dev ? libssl-dev ?