EdgePro001/quicksviewer2_stage1

quicksviewer2_stage1/
├── metadata/
│   ├── llava_recap_558k.jsonl          # 558K image-text pairs metadata
│   ├── obelics_50k_train.jsonl         # 50K multimodal documents metadata
│   └── activitynet_caption.jsonl       # ActivityNet video captions metadata
├── images/
│   ├── llava_recap_558k.tar.gz         # 13GB compressed images
│   └── obelics.tar.gz                  # 18GB compressed images
└── videos/
    └── activitynet_captions.tar.gz     # 79GB compressed videos

# Extract images
tar -xzf images/llava_recap_558k.tar.gz -C images/
tar -xzf images/obelics.tar.gz -C images/

# Extract videos
tar -xzf videos/activitynet_captions.tar.gz -C videos/

from datasets import load_dataset
from huggingface_hub import hf_hub_download
import tarfile

# Download metadata
llava_data = load_dataset("EdgePro001/quicksviewer2_stage1", data_files="metadata/llava_recap_558k.jsonl")

# Download and extract images
archive_path = hf_hub_download(
    repo_id="EdgePro001/quicksviewer2_stage1",
    filename="images/llava_recap_558k.tar.gz",
    repo_type="dataset"
)

# Extract
with tarfile.open(archive_path, 'r:gz') as tar:
    tar.extractall("./data/images/")