| --- |
| pretty_name: Modelome model and research metadata (v0.1.0) |
| tags: |
| - machine-learning |
| - model-registry |
| - provenance |
| - research |
| --- |
| |
| # Modelome dataset, v0.1.0 |
|
|
| Modelome is an evidence-linked registry of machine-learning model and research |
| metadata. This release is a snapshot from source commit `00000000000000000018-bd2b210ce6463571-41e46451164d`. Its |
| coverage is partial: the bundle represents configured sources and completed |
| work recorded at build time, not an exhaustive inventory of models or sources. |
| The bundle status is `complete`. Integrity checks: passed. |
|
|
| ## Contents |
|
|
| The bundle contains 21 entries and 74 linked |
| resources; 74 resources have no resolved artifact in this snapshot. |
| Entry resource categories are summarized below. Categories and links describe |
| recorded evidence, not independent verification that an artifact is current or |
| available. |
|
|
| | Entry resource category | Count | |
| | --- | ---: | |
| | `resource` | 63 | |
| | `weights` | 11 | |
|
|
| The manifest lists these bundled files: |
|
|
| - `entries/entries.jsonl` |
| - `entries/manifest.json` |
| - `metadata/README.md` |
| - `metadata/artifact_identifiers.parquet` |
| - `metadata/artifact_url_links.parquet` |
| - `metadata/artifacts.parquet` |
| - `metadata/export-receipt.json` |
| - `metadata/model_aliases.parquet` |
| - `metadata/model_artifacts.parquet` |
| - `metadata/model_identifiers.parquet` |
| - `metadata/model_releases.parquet` |
| - `metadata/models.parquet` |
| - `metadata/release_identifiers.parquet` |
| - `metadata/source-manifest.json` |
| - `seeds.jsonl` |
| - `seeds.jsonl.manifest.json` |
|
|
| `entries/entries.jsonl` contains entry records. `metadata/*.parquet` contains |
| projected registry tables, and `metadata/source-manifest.json` records the |
| configured sources and observed checkpoints. `seeds.jsonl` and its manifest |
| support rebuilding the entries. Parquet tables preserve IDs and source-attributed |
| relationships; they are metadata and references, not copies of linked resources. |
|
|
| ## Coverage and evidence limits |
|
|
| Source checkpoint state is reported as recorded by each adapter: |
|
|
| - `octo-checkpoints` (octo_checkpoints): checkpoint complete; pages ingested: 1; records seen: 4. |
| - `openpi-checkpoint-manifest` (openpi_checkpoint_manifest): checkpoint complete; pages ingested: 1; records seen: 10. |
| - `openvla-checkpoints` (openvla_checkpoints): checkpoint complete; pages ingested: 1; records seen: 7. |
|
|
| A checkpoint marked complete means that adapter finished its recorded scan; it |
| does not establish historical or global completeness. A configured source may |
| have no observed checkpoint. Evidence links identify source records and |
| relationships available to Modelome; they do not imply that each association, |
| license, or external URL was independently verified. Technique or method counts |
| are not inferred where the schema does not distinguish them from model entries. |
|
|
| This release contains no model weights, raw source payloads, abstracts, paper |
| full text, or copied model-card/README bodies. Follow external identifiers and |
| URLs for source material under its own terms. No single license is asserted for |
| all upstream records: users must review the terms and attribution requirements |
| that apply to each source. In particular, records derived from Papers with Code |
| archival datasets may carry attribution and share-alike obligations; arXiv |
| metadata is factual metadata and does not convey rights to paper content. |
|
|
| ## Provenance and citation |
|
|
| For reproducibility, cite the Modelome dataset repository and identify the |
| specific dataset revision you used, along with this bundle's source commit |
| `00000000000000000018-bd2b210ce6463571-41e46451164d` and Modelome version `0.1.0`. Also cite the original source |
| records used in your work. Project code and documentation: [Modelome on |
| GitHub](https://github.com/modelomics/modelome). Dataset: [Modelome on the |
| Hugging Face Hub](https://huggingface.co/datasets/modelomics/modelome). |
|
|
| Suggested citation template (fill in the Hub revision actually used): |
|
|
| > Modelome contributors. Modelome dataset, v0.1.0. Hugging Face Datasets, |
| > `modelomics/modelome`, revision `<revision>`, source commit |
| > `00000000000000000018-bd2b210ce6463571-41e46451164d`. Accessed `<date>`. |
|
|
| ## Resuming a partial collection |
|
|
| Continue from the same local Modelome store so source adapter checkpoints are |
| retained. Run `modelome build-dataset --source SOURCE --max-pages N` (repeat |
| `--source` for additional adapters) with the same `--store` and choose a new |
| `--output` directory. A page budget may produce another partial snapshot; inspect |
| the new root manifest and source manifest before using it. See the repository's |
| [`docs/dataset.md`](https://github.com/modelomics/modelome/blob/main/docs/dataset.md) |
| for build inputs, output layout, and recovery details. |
|
|
