oddadmix/laion-coco-nllb-arabic-filtered

LAION-COCO-NLLB, Arabic slice, filtered The arb_Arab captions of visheratin/laion-coco-nllb, extracted and filtered for training an Arabic captioning model. This is the exact corpus behind oddadmix/Nawah-VL-25M. split rows kept from train 753,883 878,978 (85.8%) test 14,407 14,906 (96.7%) Columns column type notes id string the source dataset's image id url string image URL, not the image. See below. caption string the arb_Arab… See the full description on the dataset page: https://huggingface.co/datasets/oddadmix/laion-coco-nllb-arabic-filtered.