tqdm
transformers<5,>=4.4
torch>=1.6.0
numpy
nltk
lingua-language-detector
sentencepiece
optimum[onnxruntime-gpu]<2,>=1.5
aiohttp<5,>=3.8.1
