metadata
configs:
- config_name: default
data_files:
- split: refusal
path: data/refusal-*
- split: non_refusal
path: data/non_refusal-*
dataset_info:
features:
- name: documents
dtype: string
- name: query
dtype: string
- name: answer
dtype: string
- name: citation_ids
dtype: string
- name: meta
dtype: string
splits:
- name: refusal
num_bytes: 51514308
num_examples: 1013
- name: non_refusal
num_bytes: 67253425
num_examples: 1232
download_size: 47461178
dataset_size: 118767733
Dataset for CReSt
This repository contains the dataset used in the paper "CReSt: A Comprehensive Benchmark for Retrieval-Augmented Generation with Complex Reasoning over Structured Documents".
📂 Dataset Overview
This dataset consists of the following two subsets:
- refusal: The subset that contains only refusal cases
- non_refusal: The subset that contains only non-refusal cases
🚀 How to Use
You can access detailed usage instructions in the CReSt GitHub repository.
Simply clone the repository and follow the steps provided in the README there:
git clone https://github.com/UpstageAI/CReSt
📜 License
This dataset is distributed under the CC-by-NC 4.0.
📝 Citation
If you use this dataset in your research, please cite:
@inproceedings{khang2025crest,
title={CReSt: A Comprehensive Benchmark for Retrieval-Augmented Generation with Complex Reasoning over Structured Documents},
author={Khang, Minsoo and Park, Sangjun and Hong, Teakgyu and Jung, Dawoon},
booktitle={TBD},
pages={TBD},
year={2025}
}