huggingface_hub==0.25.2
gradio
transformers
torch # Or tensorflow, depending on your model's backend
accelerate