torch
transformers
datasets
scikit-learn
gradio
accelerate>=0.26.0