TRIBE v2 model and weights are subject to CC-BY-NC-4.0 (research / non-commercial). See upstream:
This experiment repo does not redistribute those weights. Copy or hardlink them
from an existing cache (scripts/import-weights.sh).
Encode / render / doctor / fetch instrument vendored from
https://github.com/sw30labs/videocortex @ b34589552925338491cee8aee7087722be21c7d9
(MIT, Nicolas Cravino). VideoCortex remains an upstream reference only; this
repo does not depend on a sibling checkout at runtime and does not modify that
tree. The command deck (serve / web) is not vendored.
Pin: cortex_hit_encode.VENDOR_COMMIT.
TRIBE v2 stacks four frozen encoders. Each carries its own terms (same table as
VideoCortex NOTICE.md):
| modality | repo | note |
|---|---|---|
| text | meta-llama/Llama-3.2-3B |
Gated. Llama 3.2 Community License — accept on HuggingFace. |
| image | facebook/dinov2-large |
Apache-2.0 |
| audio | facebook/w2v-bert-2.0 |
MIT |
| video | facebook/vjepa2-vitg-fpc64-256 |
See the model card |
d'Ascoli, S., Rapin, J., Benchetrit, Y., Brooks, T., Begany, K., Raugel, J., Banville, H., & King, J.-R. (2026). A foundation model of vision, audition, and language for in-silico neuroscience. arXiv:2605.04326
Audio/video used in runs must be rights-cleared by the operator. Do not commit copyrighted media to git. Manifests may store hashes and rights basis only.