datasets >= 1.1.3
pytest<8.0.1
conllu nltk rouge-score seqeval
tensorboard
evaluate >= 0.2.0 torch accelerate
