Datasets:
File size: 2,875 Bytes
e5933de | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 | ---
configs:
- config_name: default
data_files:
- split: test
path: "MatSciBench.parquet"
dataset_info:
features:
- name: qid
dtype: string
- name: type
dtype: string
- name: question
dtype: string
- name: image
sequence: image
- name: solution
dtype: string
- name: answer
dtype: string
- name: unit
dtype: string
- name: notes
dtype: string
- name: number_of_answers
dtype: string
- name: difficulty_level
dtype: string
- name: primary_category
dtype: string
- name: Materials
dtype: string
- name: Properties
dtype: string
- name: Structures
dtype: string
- name: Fundamental Mechanisms
dtype: string
- name: Processes
dtype: string
- name: Failure Mechanisms
dtype: string
- name: source
dtype: string
- name: original_qid
dtype: string
task_categories:
- question-answering
language:
- en
tags:
- materials-science
- benchmark
- qa
size_categories:
- 1K<n<10K
---
# MatSciBench Dataset
MatSciBench is a comprehensive benchmark dataset for materials science question answering. This is the dataset for the paper [MatSciBench: Benchmarking the Reasoning Ability of Large Language Models in Materials Science](https://arxiv.org/abs/2510.12171). The dataset contains questions spanning various domains within materials science, including polymer science, crystallography, thermodynamics, and more.
For instructions on running the evaluation pipeline, see the [MatSciBench GitHub repository](https://github.com/Jun-Kai-Zhang/MatSciBench).
## Dataset Structure
The dataset is provided as a self-contained Parquet file (`MatSciBench.parquet`) with the following columns:
- `qid`: Question identifier
- `type`: Question type (NUM for numerical, etc.)
- `question`: The question text
- `image`: Associated image(s) embedded inline as image bytes; rows without images contain an empty list
- `solution`: Step-by-step solution
- `answer`: Final answer
- `unit`: Units for numerical answers
- `notes`: Additional notes
- `number_of_answers`: Number of possible answers
- `difficulty_level`: Difficulty rating
- `primary_category`: Main category classification
- Additional category columns for Materials, Properties, Structures, etc.
- `source`: Source of the question
- `original_qid`: Original question ID from source
## Citation
```bibtex
@misc{zhang2025matscibenchbenchmarkingreasoningability,
title={MatSciBench: Benchmarking the Reasoning Ability of Large Language Models in Materials Science},
author={Junkai Zhang and Jingru Gan and Xiaoxuan Wang and Zian Jia and Changquan Gu and Jianpeng Chen and Yanqiao Zhu and Mingyu Derek Ma and Dawei Zhou and Ling Li and Wei Wang},
year={2025},
eprint={2510.12171},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2510.12171},
}
```
|