Model description

Code and model is available here: https://github.com/facebookresearch/dinov2

Full paper here: https://arxiv.org/abs/2304.07193

The implementation seems fairly simple. Most layers is already implemented within transformers library (it's just a ViT). There's some changes compared to DINO (which is implemented already), such as SwiGLU and LayerScale. According to #20403, SwiGLU is already implemented, though, the original code uses xformers's SwiGLU.

DINO V2 also have a different license as listed here: https://github.com/facebookresearch/dinov2/blob/main/LICENSE
It is NonCommercial.

Open source status

• The model implementation is available
• The model weights are available

Provide useful links for the implementation

No response

If there's no issue with license, I can make a PR for the model.