A character-level encoder-decoder transformer (3+3 layers) trained on 20,532 Swedish-English Tatoeba pairs for EDAN20. It is small, so expect mistakes.