metadata
dataset_info:
- config_name: chinese
features:
- name: id
dtype: int64
- name: difficulty
dtype: string
- name: grid_image
dtype: image
- name: empty_grid_image
dtype: image
- name: key_image
dtype: image
- name: partial_0.25
dtype: image
- name: partial_0.5
dtype: image
- name: partial_0.75
dtype: image
- name: partial_grid_0.25
dtype: string
- name: partial_grid_0.5
dtype: string
- name: partial_grid_0.75
dtype: string
- name: puzzle_state
dtype: string
- name: reference_answer
dtype: string
splits:
- name: 7x7
num_bytes: 112256348
num_examples: 100
download_size: 110690847
dataset_size: 112256348
- config_name: commonsenseqa
features:
- name: id
dtype: int64
- name: difficulty
dtype: string
- name: grid_image
dtype: image
- name: empty_grid_image
dtype: image
- name: key_image
dtype: image
- name: partial_0.25
dtype: image
- name: partial_0.5
dtype: image
- name: partial_0.75
dtype: image
- name: partial_grid_0.25
dtype: string
- name: partial_grid_0.5
dtype: string
- name: partial_grid_0.75
dtype: string
- name: puzzle_state
dtype: string
- name: reference_answer
dtype: string
splits:
- name: 7x7
num_bytes: 50695997
num_examples: 50
download_size: 49635229
dataset_size: 50695997
- config_name: english
features:
- name: id
dtype: int64
- name: difficulty
dtype: string
- name: grid_image
dtype: image
- name: empty_grid_image
dtype: image
- name: key_image
dtype: image
- name: partial_0.25
dtype: image
- name: partial_0.5
dtype: image
- name: partial_0.75
dtype: image
- name: partial_grid_0.25
dtype: string
- name: partial_grid_0.5
dtype: string
- name: partial_grid_0.75
dtype: string
- name: puzzle_state
dtype: string
- name: reference_answer
dtype: string
splits:
- name: 7x7
num_bytes: 50750940
num_examples: 100
- name: 14x14
num_bytes: 144557064
num_examples: 100
download_size: 431280934
dataset_size: 195308004
- config_name: english_simple
features:
- name: id
dtype: int64
- name: difficulty
dtype: string
- name: grid_image
dtype: image
- name: empty_grid_image
dtype: image
- name: key_image
dtype: image
- name: partial_0.25
dtype: image
- name: partial_0.5
dtype: image
- name: partial_0.75
dtype: image
- name: partial_grid_0.25
dtype: string
- name: partial_grid_0.5
dtype: string
- name: partial_grid_0.75
dtype: string
- name: puzzle_state
dtype: string
- name: reference_answer
dtype: string
splits:
- name: 7x7
num_bytes: 78745737
num_examples: 100
download_size: 76652651
dataset_size: 78745737
configs:
- config_name: chinese
data_files:
- split: 7x7
path: chinese/7x7-*
- config_name: commonsenseqa
data_files:
- split: 7x7
path: commonsenseqa/7x7-*
- config_name: english
data_files:
- split: 7x7
path: english/7x7-*
- split: 14x14
path: english/14x14-*
- config_name: english_simple
data_files:
- split: 7x7
path: english_simple/7x7-*
CrossWordBench: Evaluating the Reasoning Capabilities of LLMs and LVLMs with Controllable Puzzle Generation
Citation
If find this benchmark useful, please consider citing with the following BibTex entry:
@misc{leng2025crosswordbenchevaluatingreasoningcapabilities,
title={CrossWordBench: Evaluating the Reasoning Capabilities of LLMs and LVLMs with Controllable Puzzle Generation},
author={Jixuan Leng and Chengsong Huang and Langlin Huang and Bill Yuchen Lin and William W. Cohen and Haohan Wang and Jiaxin Huang},
year={2025},
eprint={2504.00043},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2504.00043},
}