OpenLLM-Ro/ro_mwildvision

Dataset Description

m-WildVision is a multilingual multimodal LLM evaluation set covering 23 languages.

Here we provide the Romanian version of m-WildVision augmented with human curated answers. This dataset is used as a benchmark and is part of the evaluation protocol for Romanian VLMs proposed in "Înțelegi românește?" A Recipe for Romanian Vision-Language Models (Masala et al., 2026)

Citation

@article{dash2025aya,
  title={Aya vision: Advancing the frontier of multilingual multimodality},
  author={Dash, Saurabh and Nan, Yiyang and Dang, John and Ahmadian, Arash and Singh, Shivalika and Smith, Madeline and Venkitesh, Bharat and Shmyhlo, Vlad and Aryabumi, Viraat and Beller-Morales, Walter and others},
  journal={arXiv preprint arXiv:2505.08751},
  year={2025}
}

@misc{masala2026inctelegiromanecstereciperomanian,
      title={"\^{I}n\c{t}elegi Rom\^ane\c{s}te?'' A Recipe for Romanian Vision-Language Models}, 
      author={Mihai Masala and Marius Leordeanu and Mihai Dascalu and Traian Rebedea},
      year={2026},
      eprint={2605.31401},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2605.31401}, 
}