SUMMIT

• Symbolic Unified Model for Multilingual Inference and Translation

Overview

This git project consists of two parts, the python-based ML program with flask web-API and a Vue-based frontend.

Docker

To just spin up the containers, it's required to have docker and docker-compose installed on your system. Launch the containers using:

docker compose up

Prerequesites

Install git and git-lfs. After checking out the repository, run git lfs fetch and git lfs pull to download LFS files. There are no submodules to worry about.

Application

Requirements:

• Install Python and PyTorch on your system. The application was built with Python 3.12 and Cuda 12.6.

  • To make life easier with parallel python installations, the use of either venv or anaconda is recommended. Make sure that pip is available and functioning.

• Install cuda packages for your system if you have an Nvidia GPU.

• Install pip packages:

  pip install -r requirements_cpu.txt # for CPU only
  pip install -r requirements_cuda.txt # if CUDA is available on your system

  Note for developers:

  The requirements.txt file is used exclusively for building the docker image. Please do not use it. If a new package is required or updates recommended, please create new requirements.txt files using: pip freeze > requirements.txt

You can launch the application in evaluation mode using:

python src/main.py

This will also start the Web API.

Be aware that for that to work out, existing tokenizers and model files are required!

Those files should be placed into train/model and train/tokenize respectively, unless different paths are chosen in the config.json file.

To put the model into training mode, run the application using an argument:

python src/main.py train

Note: Training is 10+ times faster with a GPU

Website

The simple website is built using Vue.js and its server Vite. Both can easily be installed using bun (see below). To install bun, make sure you have nodeJS (current LTS, v22.13.1) installed. Multiple parallel node versions can be managed using nvm. After installation of nvm (see installation instructions on the site), make sure to install the current LTS version of node:

nvm ls-remote
nvm install --lts
nvm use --lts

Install bun either via node or directly on the system using the executable, the latter is recommended if using nvm as otherwise, bun might only be installed to the currently selected node instance.

Once all requirements are installed, install the node modules using:

bun install

The website can then be launched using:

bun dev

Hint: if this should throw errors, check the currently used node version using node --version