Datasets:

Modalities:
Text
Formats:
parquet
Size:
< 1K
ArXiv:
License:
e1-verifiable / README.md
emendes3's picture
Update README.md
604cce5 verified
|
Raw
History Blame Contribute Delete
894 Bytes
---
license: mit
---
# e1-verifiable
e1-verifiable contains challenging problems from the American Invitational Mathematics Examination (AIME), i.e., pass@32 = 0 for Qwen2.5-7B-Instruct. The solutions were sourced from [Yue et al.'s](https://arxiv.org/pdf/2412.08819) scrape of the [AoPS Wiki](https://artofproblemsolving.com/wiki/index.php?title=AIME_Problems_and_Solutions).
This dataset was collected in this work: [Making Expert Reasoning Learnable with Self-Distillation](https://arxiv.org/abs/2602.02405). If you use this dataset, please cite this paper:
```
@inproceedings{
mendes2026selfdistill,
title={Making Expert Reasoning Learnable with Self-Distillation},
author={Mendes, Ethan and Park, Jungsoo and Ritter, Alan},
booktitle={Forty-third International Conference on Machine Learning},
year={2026},
url={https://openreview.net/forum?id=JG6f02X29a}
}
```