Dataset Viewer
Duplicate
The dataset viewer is not available for this split.
The size of the content of the first rows (6260380 B) exceeds the maximum supported size (200000 B) even after truncation. Please report the issue.
Error code:   TooBigContentError

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

OMGM InfoSeek candidate-cache metadata

This public artifact contains metadata and scripts used to build the OMGM InfoSeek candidate-side embedding cache in our reproduction experiments.

Included

  • metadata/index.json: URL-to-embedding offset/count mapping and cache schema.
  • metadata/inventory.json: sanitized article inventory containing only URL, title, offset, and count. Machine-local image paths were removed.
  • metadata/progress.json: completion record and measured cache-build wall time.
  • scripts/build_it2it_candidate_cache.py: cache construction script.
  • scripts/sanitize_inventory.py: transformation used to remove local paths.

Not included

  • embeddings.npy (about 17 GB)
  • Wikipedia image pixels
  • model checkpoints
  • private filesystem paths, credentials, or access tokens

Consequently, index.json describes the complete cache but cannot be used for inference without regenerating embeddings.npy using the supplied script and the required upstream assets.

Provenance

The metadata was produced while reproducing OMGM on InfoSeek. It is an unofficial derived research artifact, not an official OMGM release. Users must obtain upstream datasets, images, and model checkpoints under their respective terms and should cite OMGM, InfoSeek, Wikipedia, and the relevant BLIP-2/LAVIS components. No upstream image pixels or checkpoints are redistributed here.

The repository uses license: other because upstream OMGM materials in the checked source tree did not include a top-level license. Publication of this metadata does not grant rights to upstream assets.

Downloads last month
11