einops numpy torchaudio==2.8.0 transformers safetensors accelerate flashpack==0.1.2 scikit-image>=0.25.2 av tqdm pillow scipy>=1.14 flash-attn-3 @ https://huggingface.co/alexnasa/flash-attn-3/resolve/main/128/flash_attn_3-3.0.0b1-cp39-abi3-linux_x86_64.whl bitsandbytes