kilian-group/phantom-reasoning

---
license: cc-by-4.0
---

# Phantom Reasoning

[](https://arxiv.org/abs/2603.02091)
[](https://github.com/kilian-group/phantom-reasoning)

<!-- Provide a quick summary of the dataset. -->

We provide `.zip` files for all training and evaluation data for the paper. After downloading the source code, and unzipping files under `data/`, the dataloader will load them correctly.

Evaluation benchmarks (`hp500, 2wiki500, msq500, cofca500, synthrm500`) are random subsets of 500 samples from publicly available datasets. Please attribute use to their work (see the paper for references).

## Citation

```bibtex
@inproceedings{kabra2026learning,
title={{Learning from Synthetic Data Improves Multi-hop Reasoning}},
author={Kabra, Anmol and Gong, Albert and Yin, Yilun and Stankevi{\v{c}}i{\=u}t{\.e}, Kamil{\.e} and Go, Dongyoung and Luo, Katie Z and Lee, Johann and Gomes, Carla P and Weinberger, Kilian Q},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://arxiv.org/abs/2603.02091}
}
```