--- pretty_name: MODELOME public metadata tags: - machine-learning - model-registry - provenance - research --- # MODELOME public metadata This is a provenance-preserving, metadata-only export of the MODELOME registry. It records model identities, source-attributed artifact links, and trained-release identifiers. The source registry commit is 00000000000000000018-bd2b210ce6463571-41e46451164d (generation 18). ## Included files - artifact_identifiers.parquet: 21 rows. - artifact_url_links.parquet: 32 rows. - artifacts.parquet: 21 rows. - model_aliases.parquet: 41 rows. - model_artifacts.parquet: 21 rows. - model_identifiers.parquet: 21 rows. - model_releases.parquet: 21 rows. - models.parquet: 21 rows. - release_identifiers.parquet: 21 rows. model_artifacts.parquet joins model IDs to their current source artifacts. artifact_url_links.parquet joins a current artifact to a discovered external URL and, when present in this export, the matching target artifact ID. Together with artifacts.parquet and identifiers, those tables support paper/code/model and model/release reconstruction without claiming that a URL is a binary payload. model_releases.parquet may include weight_files_json, a JSON array of safe relative filenames declared by a Hugging Face model record. weight_files_complete says whether the revision's file listing finished; false means the array contains only observed files. Other release metadata is omitted. ## Deliberately excluded This release does not include source-record raw payloads, paper abstracts or full text, model-card bodies, README bodies, source HTML, or unreviewed evidence values. Follow the external URLs and identifiers for those materials under their own terms. ## Source attribution and rights source-manifest.json records configured sources and observed source checkpoints, including configured sources with no observed checkpoint. The checkpoint's complete flag and counts are reported only when that checkpoint exists; they do not claim unrun sources were ingested. A true complete flag means the adapter finished its checkpointed scan; it does not claim exhaustive historical or global coverage. Source-level terms remain in force. In particular, derivative users must preserve attribution and applicable share-alike obligations for records derived from the Papers with Code archival datasets. The arXiv snapshot contributes factual metadata only; paper-specific licenses remain attached to the original records and paper content is not included. Contributing source checkpoints: - octo-checkpoints (octo_checkpoints) - openpi-checkpoint-manifest (openpi_checkpoint_manifest) - openvla-checkpoints (openvla_checkpoints)