dpo_zh_demo:
  path: HuggingFaceH4/orca_dpo_pairs
  split: train_prefs
  converter: pair