language:
- it
This is a machine-translated version of The Cauldron for the Italian language.
Out of the 50 tasks, only a subset of 15 tasks which do not lose their meaning after machine translation are considered (specifically, tasks which focus on the textual contents of images are removed). Out of these 15 tasks, the first 10k rows are selected for machine translation. Question-Answer pairs that were not correctly translated were discarded.
Image paths are formatted using the following strategy:
{task-name}/images/{row_number}_{image_number}
Where {task-name} is the name of the task from the original dataset, {row_number} is the row number in the original dataset and {image_number} the index of the image (in case of tasks where there can be multiple images as input).
Citation
If you use this dataset in your research please cite the following
@inproceedings{musacchioLLaVANDiNO,
title={LLaVA-NDiNO: Empowering LLMs with Multimodality for the Italian Language},
author={Musacchio, Elio and Siciliani, Lucia and Basile, Pierpaolo and Semeraro, Giovanni},
booktitle={Proceedings of the Eighth Workshop on Natural Language for Artificial Intelligence (NL4AI 2024) co-located with 23th International Conference of the Italian Association for Artificial Intelligence (AI*IA 2024)},
year={2024}
}
@misc{laurençon2024matters,
title={What matters when building vision-language models?},
author={Hugo Laurençon and Léo Tronchon and Matthieu Cord and Victor Sanh},
year={2024},
eprint={2405.02246},
archivePrefix={arXiv},
primaryClass={cs.CV}
}