torch>=2.10
torchvision
transformers==5.17.0
accelerate
huggingface_hub
safetensors
numpy
Pillow
opencv-python-headless
soundfile
librosa