README.md

---
task_categories:
- text-ranking
tags:
- creative-writing
- llm-evaluation
- preference-alignment
- reward-modeling
- benchmark
- reddit
dataset_info:
  features:
  - name: prompt
    dtype: string
  - name: chosen_story
    dtype: string
  - name: rejected_story
    dtype: string
  - name: chosen_timestamp
    dtype: timestamp[ns]
  - name: rejected_timestamp
    dtype: timestamp[ns]
  - name: chosen_upvotes
    dtype: int64
  - name: rejected_upvotes
    dtype: int64
  splits:
  - name: train
    num_bytes: 276261399
    num_examples: 43827
  download_size: 172500713
  dataset_size: 276261399
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# LitBench: A Benchmark and Dataset for Reliable Evaluation of Creative Writing

LitBench is the first standardized benchmark and paired dataset for reliable evaluation of creative writing generated by large language models (LLMs). It addresses the challenge of evaluating open-ended narratives, which lack ground truths. The dataset comprises a held-out test set of 2,480 debiased, human-labeled story comparisons drawn from Reddit and a 43,827-pair training corpus of human preference labels. LitBench facilitates benchmarking zero-shot LLM judges and training reward models for creative writing verification and optimization.

**Paper:** [LitBench: A Benchmark and Dataset for Reliable Evaluation of Creative Writing](https://huggingface.co/papers/2507.00769)

**Project Page (Hugging Face Collection):** https://huggingface.co/collections/SAA-Lab/litbench-68267b5da3aafe58f9e43461

### Sample Usage

You can load the dataset using the Hugging Face `datasets` library:

```python
from datasets import load_dataset

dataset = load_dataset("SAA-Lab/LitBench")

# Access the training split
train_dataset = dataset["train"]

# Print the first example
print(train_dataset[0])
```

If you are the author of any comment in this dataset and would like it removed, please contact us and we will comply promptly.