This guide shows how to download split archives from Hugging Face, merge them into a single ZIP, extract, and place the images into the LibMoE data tree:
libmoe/
└── data/
├── image_onevision/
├── coco/
│ └── train2017/
├── gqa/
│ └── images/
├── ocr_vqa/
│ └── images/
├── textvqa/
│ └── train_images/
└── vg/
├── VG_100K/
└── VG_100K_2/
Tested on Linux… See the full description on the dataset page:
https://huggingface.co/datasets/DavidNguyen/ocr_vqa.
