translate_dataset

A project for translating datasets to Danish

To do

The code runs. Quality check the output
Select best translation model: Translate the Skolegpt-instruct subset of OpenOrca from English to Danish with several candidate models (e.g. Llama 70b, Gemma 3 27b, Qwen3 32b etc) and select the model that produces the best translations (eg the one with the best Comet scores or the shortest Libre distance) where the Skolegpt-instruct is the reference translations.
Translate MS Marco using the run.py script and the selected translation model

To set up a server with the pre-requisite software, run the setup_server.sh script like so:

source setup_server.sh

Note that you will be prompted to enter a Hugging Face access token in order to access any gated repo that your code uses.

Please upload any final datasets to the DDSC organization on Hugging Face. Open an issue to get access :)

Name		Name	Last commit message	Last commit date
Latest commit History 32 Commits
.gitignore		.gitignore
.python-version		.python-version
README.md		README.md
pyproject.toml		pyproject.toml
run.py		run.py
setup_server.sh		setup_server.sh
translators.py		translators.py
uv.lock		uv.lock