** DPO:
python -u train.py \
  model=gpt2 \
  model.name_or_path=model_hub/gpt2_120M/ \
  loss=dpo \
  base_data_dir=data \
  datasets='["Ultrachat200k/DPO"]'

** WSPIN: python -u train.py model=gpt2 model.name_or_path=model_hub/gpt2_120M/
