README.md

# Prepared Dataset Archives

This repository stores a prepared split as tar.zst archives to avoid
millions of small files in the Hugging Face repository.

Extraction:

```bash
mkdir -p data
for f in *.tar.zst; do
  tar --use-compress-program=zstd -xf "$f" -C data
done
sha256sum -c SHA256SUMS
```

After extraction, the layout is:

```text
data/
  dataset.json
  dataset_stats.json
  jsonl/
  frames/
  vision_cache/
```