Models

The default is --dino-version auto --size base. Automatic mode uses the selected DINOv3 model when it is already cached or can be accessed with the active Hugging Face account. Otherwise it reports the fallback and uses the corresponding DINOv2 model.

Explicit DINOv2 selection uses --dino-version 2. DINOv2 supports small, base, large, and max; max selects DINOv2 Giant. The --arch option is ignored for DINOv2. Automatic tiny and huge selections fall back to DINOv2 Small and Giant respectively.

DINOv3 presets are selected with --dino-version 3 and use both --arch and --size. Smaller models are faster and need less memory. An arbitrary compatible Hugging Face model can be supplied with --model, which overrides the version, architecture, and size flags.

Getting access to DINOv3

The DINOv3 repositories are gated. Before the first run:

Sign in to Hugging Face and open the DINOv3 ViT-B/16 model page.
Review and accept Meta’s DINOv3 license and agree to share the requested contact information. Approval is usually automatic, but access can take several minutes (often 5–15 minutes) to propagate.
Authenticate locally and verify the active account:
```
hf auth login
hf auth whoami
```

The login command uses a browser flow and stores a token locally. On a server or in another non-interactive environment, create a read token in Hugging Face token settings and provide it through the environment:

export HF_TOKEN=hf_your_token_here

Do not commit the token or embed it in scripts. The token must belong to the same Hugging Face account that accepted the DINOv3 terms. Model weights are governed by the DINOv3 license rather than imcluster’s Apache License 2.0.

Inference uses pooled, normalized embeddings. --device auto prefers CUDA, then Apple MPS, then CPU. Adjust --batch-size to fit available memory.