README.md

metadata

pretty_name: LitBench Preference Pairs (chat format)
dataset_name: CarlOwOs/LitBench-Processed
task_categories:
  - text-generation
  - preference-modeling
tags:
  - rlhf
  - pairwise-preference
  - conversations
  - literature
language:
  - en
size_categories:
  - n<1M

LitBench Preference Pairs (chat format)

This dataset consolidates the LitBench train and test preference pairs into one repository with two splits: train and test. Each example contains two chat-style conversations under keys chosen and rejected, plus numeric scores score_chosen and score_rejected.

Last updated: 2025-10-04 05:15 UTC

Splits

• train: 43827 examples
• test: 0 examples

Features

• chosen: list of role/content messages representing the preferred conversation
• rejected: list of role/content messages representing the alternative conversation
• score_chosen: integer upvote score of the chosen story
• score_rejected: integer upvote score of the rejected story

Loading

from datasets import load_dataset

ds = load_dataset("CarlOwOs/LitBench-Processed")
print(ds)

Citation

If you use this dataset, please also acknowledge the original LitBench sources.