Upload README.md with huggingface_hub
Browse files
README.md
CHANGED
|
@@ -35,8 +35,6 @@ dataset_info:
|
|
| 35 |
dtype: bool
|
| 36 |
- name: other_instructions
|
| 37 |
dtype: string
|
| 38 |
-
- name: blacklist_packages
|
| 39 |
-
sequence: string
|
| 40 |
- name: full_text
|
| 41 |
dtype: string
|
| 42 |
splits:
|
|
@@ -56,3 +54,52 @@ configs:
|
|
| 56 |
- split: tasks
|
| 57 |
path: tasks.jsonl
|
| 58 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
dtype: bool
|
| 36 |
- name: other_instructions
|
| 37 |
dtype: string
|
|
|
|
|
|
|
| 38 |
- name: full_text
|
| 39 |
dtype: string
|
| 40 |
splits:
|
|
|
|
| 54 |
- split: tasks
|
| 55 |
path: tasks.jsonl
|
| 56 |
---
|
| 57 |
+
|
| 58 |
+
# ReplicationBench
|
| 59 |
+
**arXiv**: [ReplicationBench: Can AI Agents Replicate Astrophysics Research Papers?](https://arxiv.org/abs/2510.24591)
|
| 60 |
+
|
| 61 |
+
**GitHub**: [https://github.com/Christine8888/replicationbench-release](https://github.com/Christine8888/replicationbench-release)
|
| 62 |
+
|
| 63 |
+
## Dataset Description
|
| 64 |
+
|
| 65 |
+
The ReplicationBench dataset contains 114 astrophysics research replication tasks spanning 20 research papers. The dataset includes:
|
| 66 |
+
- Original and masked manuscript text
|
| 67 |
+
- Metadata (title, abstract, publication info, etc.)
|
| 68 |
+
- Pointers to datasets and dataset access instructions
|
| 69 |
+
- Additional specifications from the authors
|
| 70 |
+
- Execution requirements
|
| 71 |
+
- Detailed descriptions and grading guidelines for each task
|
| 72 |
+
|
| 73 |
+
## Usage
|
| 74 |
+
|
| 75 |
+
```python
|
| 76 |
+
from datasets import load_dataset
|
| 77 |
+
|
| 78 |
+
# Load papers
|
| 79 |
+
papers_ds = load_dataset("ChristineYe8/replicationbench", split="papers")
|
| 80 |
+
|
| 81 |
+
# Load tasks
|
| 82 |
+
tasks_ds = load_dataset("ChristineYe8/replicationbench", split="tasks")
|
| 83 |
+
```
|
| 84 |
+
|
| 85 |
+
You can load the dataset from HuggingFace into native ReplicationBench format using [this script](https://github.com/Christine8888/replicationbench-release/blob/main/src/dataset/hf/load_from_hf.py). However, if using RB's native formats, we recommend using the native data loading instead, described [here](https://github.com/Christine8888/replicationbench-release).
|
| 86 |
+
|
| 87 |
+
## Citation
|
| 88 |
+
|
| 89 |
+
If you use ResearchBench in your research, please cite:
|
| 90 |
+
|
| 91 |
+
```bibtex
|
| 92 |
+
@misc{ye2025replicationbenchaiagentsreplicate,
|
| 93 |
+
title={ReplicationBench: Can AI Agents Replicate Astrophysics Research Papers?},
|
| 94 |
+
author={Christine Ye and Sihan Yuan and Suchetha Cooray and Steven Dillmann and Ian L. V. Roque and Dalya Baron and Philipp Frank and Sergio Martin-Alvarez and Nolan Koblischke and Frank J Qu and Diyi Yang and Risa Wechsler and Ioana Ciuca},
|
| 95 |
+
year={2025},
|
| 96 |
+
eprint={2510.24591},
|
| 97 |
+
archivePrefix={arXiv},
|
| 98 |
+
primaryClass={cs.CL},
|
| 99 |
+
url={https://arxiv.org/abs/2510.24591},
|
| 100 |
+
}
|
| 101 |
+
```
|
| 102 |
+
|
| 103 |
+
## License
|
| 104 |
+
|
| 105 |
+
MIT License
|