cairo2000/dinov3-vitb16-pretrain-lvd1689m-embed

DINOv3 ViT-B/16 embedding endpoint (custom handler)

Deploy this repo as a Hugging Face Inference Endpoint (Task = Custom).

Weights are loaded from gated facebook/dinov3-vitb16-pretrain-lvd1689m at startup (pooler_output + L2-norm → 768-d).

Required: HF token for gated Meta model

1. Accept access on https://huggingface.co/facebook/dinov3-vitb16-pretrain-lvd1689m
2. In the Inference Endpoint → Settings / Secrets, add:
   • HF_TOKEN = your Hugging Face write/read token (account that accepted the license)
3. Optional env: DINO_MODEL_ID=facebook/dinov3-vitb16-pretrain-lvd1689m
4. Restart / update the endpoint after changing secrets

Without HF_TOKEN the container gets 401 gated repo on boot.

Request

{"inputs": {"images": ["data:image/jpeg;base64,..."]}, "normalize": true}

Response

image_embeddings, dim, timings_ms