We preprocess the dataset using our standard prompt, chosen, rejected format.
Download the tldr_preference.py from the
https://huggingface.co/datasets/vwxyzjn/hh-rlhf-trl-style/tree/0.1.0.
Run python examples/datasets/tldr_preference.py --debug --push_to_hub
