Dataset Card for MS-Bench

This is the official dataset for MS-Bench: Evaluating LMMs in Ancient ManuScript Study with Dunhuang as Case Study. Please refer to the GitHub page for more details: https://github.com/ianeong/MS-Bench.

Dataset Details

Dataset Description

• Curated by: Yuqing Zhang, Yue Han, Shuanghe Zhu, Haoxiang Wu, Hangqi Li
• Funded by: Zhejiang University
• Shared by: Zhejiang University
• Language(s) (NLP): Chinese
• License: MIT

Dataset Sources [optional]

• Repository: https://github.com/ianeong/MS-Bench
• Paper [optional]: https://openreview.net/forum?id=mPznBXZSWf
• Demo [optional]: https://neurips.cc/virtual/2025/poster/121478 (To Be Updated)

Uses

Direct Use

Please contact paper authors for dataset direct use.

Out-of-Scope Use

Please refer to paper.

Dataset Structure

We provide each task as a zip file with json annoatation alongside each task.

Dataset Creation

Curation Rationale

We detail in paper about dataset curation process, please refer to: https://openreview.net/forum?id=mPznBXZSWf.

Source Data

Data Collection and Processing

We detail in paper about dataset collection and preprocessing.

Who are the source data producers?

Zhejiang University.

Bias, Risks, and Limitations

Please refer to paper for discusions on limitations.

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Citation [optional]

BibTeX:

To Be Updated.

APA:

To Be Updated.

Dataset Card Authors [optional]

Yuqing Zhang

Dataset Card Contact

[email protected]