Datasets:
Upload README.md with huggingface_hub
Browse files
README.md
ADDED
|
@@ -0,0 +1,68 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
task_categories:
|
| 4 |
+
- question-answering
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
tags:
|
| 8 |
+
- math
|
| 9 |
+
- number-sense
|
| 10 |
+
- benchmark
|
| 11 |
+
- shortcuts
|
| 12 |
+
- numerical-reasoning
|
| 13 |
+
size_categories:
|
| 14 |
+
- 1K<n<10K
|
| 15 |
+
---
|
| 16 |
+
|
| 17 |
+
# SenseMath: Evaluating Number Sense in Large Language Models
|
| 18 |
+
|
| 19 |
+
**SenseMath** is a controlled benchmark for measuring whether LLMs can exploit number-sense shortcuts.
|
| 20 |
+
|
| 21 |
+
## Dataset Description
|
| 22 |
+
|
| 23 |
+
- **1,600 item families** across 8 categories and 4 digit scales
|
| 24 |
+
- **3 variants per family**: strong-shortcut, weak-shortcut, control
|
| 25 |
+
- **4,800 total items**
|
| 26 |
+
- **Categories**: Magnitude Estimation, Structural Shortcuts, Relative Distance, Cancellation, Compatible Numbers, Landmark Comparison, Equation Reasoning, Option Elimination
|
| 27 |
+
- **Digit scales**: d=2, 4, 8, 16
|
| 28 |
+
|
| 29 |
+
## Files
|
| 30 |
+
|
| 31 |
+
| File | Description |
|
| 32 |
+
|------|-------------|
|
| 33 |
+
| `data/sensemath_v2_d2.json` | 400 families, 2-digit operands |
|
| 34 |
+
| `data/sensemath_v2_d4.json` | 400 families, 4-digit operands |
|
| 35 |
+
| `data/sensemath_v2_d8.json` | 400 families, 8-digit operands |
|
| 36 |
+
| `data/sensemath_v2_d16.json` | 400 families, 16-digit operands |
|
| 37 |
+
| `data/judge_j1.json` | J1 task: shortcut recognition (251 items) |
|
| 38 |
+
| `data/judge_j2.json` | J2 task: strategy identification (80 items) |
|
| 39 |
+
| `data/judge_j3.json` | J3 task items |
|
| 40 |
+
|
| 41 |
+
## Usage
|
| 42 |
+
|
| 43 |
+
```python
|
| 44 |
+
from datasets import load_dataset
|
| 45 |
+
ds = load_dataset("DaydreamerMZM/SenseMath", split="train")
|
| 46 |
+
|
| 47 |
+
# Or load directly
|
| 48 |
+
import json
|
| 49 |
+
with open("data/sensemath_v2_d4.json") as f:
|
| 50 |
+
families = json.load(f)
|
| 51 |
+
```
|
| 52 |
+
|
| 53 |
+
## Citation
|
| 54 |
+
|
| 55 |
+
```bibtex
|
| 56 |
+
@article{zhuang2025sensemath,
|
| 57 |
+
title={SenseMath: Evaluating Number Sense in Large Language Models},
|
| 58 |
+
author={Zhuang, Haomin and Wang, Xiangqi and Shen, Yili and Cheng, Ying and Zhang, Xiangliang},
|
| 59 |
+
journal={arXiv preprint arXiv:XXXX.XXXXX},
|
| 60 |
+
year={2025}
|
| 61 |
+
}
|
| 62 |
+
```
|
| 63 |
+
|
| 64 |
+
## Links
|
| 65 |
+
|
| 66 |
+
- **Paper**: [arXiv](https://arxiv.org/abs/XXXX.XXXXX)
|
| 67 |
+
- **Code**: [GitHub](https://github.com/zhmzm/SenseMath)
|
| 68 |
+
- **Project Page**: [zhmzm.github.io/SenseMath](https://zhmzm.github.io/SenseMath/)
|