On behalf of our colleague, Laurent Besacier:

hi everyone

i thought this info could be useful for scientists working on NMT for low resource languages we recently released  SMaLL-100 model, a Shallow Multilingual MT Model for Low-Resource Languages it is a distilled version of the large 12B MTM-100 model released by Meta

you can find out more here: https://arxiv.org/abs/2210.11621

but the reason why i want to share this to SIGUL & EAMT community is because we provide models (which are a good pre-trained model to develop MT for low resource language pair) and also a demo platform to access MT for those 10,000 language pairs !

Models: https://huggingface.co/alirezamsh/small100 Online MT demo: https://huggingface.co/spaces/alirezamsh/small100 (still a bit slow because currently running on 2 v*CPU* - 16GB RAM)

Best regards

Laurent Besacier Naver Labs Europe