numpy
transformers==4.28.0
sentencepiece
tokenizers
accelerate