accelerate
evaluate
datasets >= 1.8.0
sentencepiece != 0.1.92
scipy
scikit-learn
protobuf
rouge-score
nltk
py7zr
torch >= 1.9.0
torch-ort
