Datasets:

Modalities:
Text
Formats:
csv
ArXiv:
Libraries:
Datasets
Dask
Parrot-dataset / README.md
liyang's picture
Update README.md
6dfc627 verified

Description

The Parrot dataset is a multilingual, multimodal dataset comprising two parts: the multimodal training datasets sharegpt-4v-ar, sharegpt-4v-pt, sharegpt-4v-ru, sharegpt-4v-tr, and sharegpt-4v-zh, as well as the multimodal evaluation benchmarks MMBench and MMMB. For detailed information about the dataset, please refer to: Parrot. For the images in the ShareGPT and MMBench datasets, you can refer to the original datasets to obtain them. Due to translation and review processes, the number of data points in other translated languages in mmbench will be fewer than in the original English dataset (each language will have more than 95%). The images in the MMMB dataset have been encoded in base64, so they need to be decoded from base64 for use.

License

The dataset is released under CC BY-NC-SA 4.0. The data is released for non-commercial research purposes only.

Declaration

Data Sources:

Please contact us if you believe any data infringes upon your rights, and we will remove it.