torch
pillow
requests
tyro
jsonargparse
tqdm
datasets
huggingface-hub
trl
torchvision
wandb
pyyaml
# transformers
git+https://github.com/huggingface/transformers.git@e39b6c1c7cdc890b6849b8c9de545fc9590ba871  # >=4.45.0, chat templates for multimodal models
