decord peft pandas matplotlib loguru sentencepiece dashscope ftfy diffusers opencv-python moviepy torchvision==0.23.0 torchaudio==2.8.0 transformers==4.57.6 tokenizers accelerate tqdm imageio[ffmpeg] easydict imageio-ffmpeg numpy>=1.23.5,<2 hydra-core iopath pytest pillow gradio-modal onnxruntime-gpu librosa fvcore 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