---
task_categories:
- text-retrieval
dataset_info:
  features:
  - name: query
    dtype: string
  - name: image_filename
    dtype: string
  - name: image
    dtype: image
  - name: text
    dtype: string
  splits:
  - name: train
    num_bytes: 2917901918
    num_examples: 6122
  download_size: 2867797655
  dataset_size: 2917901918
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# MP-DocVQA

`MP-DocVQA` is one of the 11 retrieval benchmarks used in **RetrievalRouter: Joint Modality and
Architecture Selection for Document Retrieval** (EMNLP 2026). Each record pairs a rendered page
image, a query, and the page's extracted text, supporting both text-based and multimodal
retrieval evaluation.

- 📄 Paper: https://arxiv.org/pdf/2608.25625
- 💻 Code: https://github.com/emrekuruu/retrieval-router
- 🤗 Collection: https://huggingface.co/collections/emrekuruu/retrieval-router

**Source benchmark:** MMDocIR (Dong et al., 2025). This repository repackages that benchmark for the
RetrievalRouter experiments; if you use it, please cite the original source as well.

## Citation

```bibtex
@misc{kuru2026retrievalrouterjointmodalityarchitecture,
      title={RetrievalRouter: Joint Modality and Architecture Selection for Document Retrieval}, 
      author={Emre Kuru and Mehmet Onur Keskin and Reza Farahbakhsh and Noel Crespi},
      year={2026},
      eprint={2608.25625},
      archivePrefix={arXiv},
      primaryClass={cs.IR},
      url={https://arxiv.org/abs/2608.25625}, 
}

@misc{dong2025mmdocir,
  title={MMDocIR: Benchmarking Multi-Modal Retrieval for Long Documents},
  author={Dong, Kuicai and Chang, Yujing and Goh, Xin Deik and Li, Dexun and Tang, Ruiming and Liu, Yong},
  year={2025},
  eprint={2501.08828},
  archivePrefix={arXiv},
  primaryClass={cs.IR}
}
```