transformers>=4.40
datasets
tokenizers
torch
scikit-learn
numpy