torch>=2.4
torchvision
transformers>=5.0
peft>=0.18
datasets
soundfile
librosa
numpy
wandb
accelerate