datasets >= 1.13.3,<2.20.0 # Temporary upper version
pytest<8.0.1
conllu nltk rouge-score seqeval
tensorboard
evaluate >= 0.2.0 torch
accelerate
