---
license: apache-2.0
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
dataset_info:
  features:
  - name: id
    dtype: string
  - name: contents
    dtype: string
  - name: title
    dtype: string
  - name: wikipedia_id
    dtype: string
  splits:
  - name: train
    num_bytes: 18038881943
    num_examples: 35678076
  download_size: 10150820540
  dataset_size: 18038881943
language:
- en
---

# KILT Corpus

This dataset contains approximately 36 million Wikipedia passages from the "[Multi-task retrieval for knowledge-intensive tasks](https://arxiv.org/pdf/2101.00117)" paper. It is also the retrieval corpus used in the paper [Chain-of-Retrieval Augmented Generation](https://arxiv.org/pdf/2501.14342).

## Fields

* `id`: A unique identifier for each passage.
* `title`: The title of the Wikipedia page from which the passage originates.
* `contents`: The textual content of the passage.
* `wikipedia_id`: The unique identifier for the Wikipedia page, used for KILT evaluation.

## How to Load the Dataset

You can easily load this dataset using the `datasets` library from Hugging Face. Make sure you have the library installed (`pip install datasets`).

```python
from datasets import load_dataset

ds = load_dataset('corag/kilt-corpus', split='train')

# You can inspect the dataset structure and the first few examples:
print(ds)
print(ds[0])
```

## References

```
@article{maillard2021multi,
  title={Multi-task retrieval for knowledge-intensive tasks},
  author={Maillard, Jean and Karpukhin, Vladimir and Petroni, Fabio and Yih, Wen-tau and O{\u{g}}uz, Barlas and Stoyanov, Veselin and Ghosh, Gargi},
  journal={arXiv preprint arXiv:2101.00117},
  year={2021}
}

@article{wang2025chain,
  title={Chain-of-Retrieval Augmented Generation},
  author={Wang, Liang and Chen, Haonan and Yang, Nan and Huang, Xiaolong and Dou, Zhicheng and Wei, Furu},
  journal={arXiv preprint arXiv:2501.14342},
  year={2025}
}
```