VennLicenses and attribution#
Venn code is MIT-licensed; see LICENSE and NOTICE. Reader weights, datasets and trained heads have separate terms. The code license does not relicense them.
| Component | License | Notes |
|---|---|---|
| Venn code | MIT | LICENSE and NOTICE accompany source downloads |
| EmbeddingGemma 2 reader | Apache-2.0 | Google DeepMind. Downloaded separately; read the upstream model card before downloading |
| NumPy | BSD, with MIT, Zlib and CC0 components | Runtime dependency |
| PyTorch | BSD-3-Clause | Runtime dependency |
| FAISS | MIT | Runtime dependency |
| FastAPI | MIT | Runtime dependency |
| Uvicorn | BSD-3-Clause | Runtime dependency |
| orjson | Apache-2.0 or MIT | Runtime dependency |
| safetensors | Apache-2.0 | Runtime dependency |
| Hugging Face Hub | Apache-2.0 | Runtime dependency |
| Freedoom assets in the arena video | Freedoom license | License notice and credits accompany the recording; no commercial Doom assets are used |
| ViZDoom | Own license | Optional. The arena example uses an installed engine and does not bundle one; custom levels keep their own terms |
| Evaluation dataset texts | Per dataset | Fetched from pinned original sources at run time; not bundled |
| Precedent Desk embedding snapshots | Banking77: CC BY 4.0; CLINC: CC BY 3.0 | Separately distributed reproduction inputs; see sources and attribution |
Optional dependencies retain their own licenses and notices; inspect the exact installed distributions, including bundled and transitive components. Alternative image readers are selected and downloaded separately under their own terms.
Venn-1 head redistribution#
Redistribution of the published venn-1 head weights is permitted. The project
owner confirmed this on November 7, 2026. The Hub card records this permission
under a custom license notice; the MIT code license is separate.
See the Hub model card
for the actual grant. This page does not expand that grant to unspecified
use, modification or commercial-use rights.
The head was trained on CLINC, MASSIVE, Banking77, DBpedia, Yahoo Answers and AG News. Source attributions and the historical review are retained in the license audit. The embedding reader and datasets keep their upstream terms.