midah/model-dataset-licenses

# HF Datasets License Catalog

Catalog of ~885k Hugging Face datasets with their licenses, built from a single pass over `list_datasets()` (license from tags). No per-dataset card fetches.

**Related:** [midah/hf-dataset-licenses](https://huggingface.co/datasets/midah/hf-dataset-licenses) (SPDX 700+ licenses) · [modelbiome/ai_ecosystem](https://huggingface.co/datasets/modelbiome/ai_ecosystem) (1.86M models — figures 05–07 use model↔dataset links)

## Dataset

- **Repo**: [midah/hf-datasets-licenses](https://huggingface.co/datasets/midah/hf-datasets-licenses)

## Schema

| Field | Type | Description |
|-------|------|--------------|
| dataset_id | string | HF dataset ID (e.g. `org/name`) |
| license | string \| null | License from tags (`license:xxx`), null if absent |
| author | string | Dataset author |
| downloads | int | Total downloads |
| likes | int | Likes |
| private | bool | Private repo |
| gated | bool | Gated access |
| last_modified | string | ISO timestamp |

## Usage

```python
from datasets import load_dataset

ds = load_dataset("midah/hf-datasets-licenses")
# Filter by license
mit = ds["train"].filter(lambda x: x["license"] == "mit")
```

## Figures


Model ↔ dataset relationships from [modelbiome/ai_ecosystem](https://huggingface.co/datasets/modelbiome/ai_ecosystem): top cross-license pairs, conditional dataset mix per model license, and consistency.

## Source

Built by `scripts/build_hf_datasets_licenses.py` from the Hugging Face Hub API. License from tags only; no per-dataset README parsing.

## License

CC0-1.0.