DINOv2
Benchmarks
| Task Name | Dataset Name | SOTA Result | Trend | |
|---|---|---|---|---|
| Interpretability and Faithfulness Evaluation | DINOv2 ViT-B/14 tokens | LLM Rank10 | 22 | |
| Computational Efficiency | DINOv2 ViT-B/14 (layer 11) | Latency (ms/batch)380.83 | 10 | |
| Dictionary Learning Stability and Geometry Evaluation | DINOv2-B/14 activations (three seeded token datasets) | Cross-Init Similarity93.34 | 9 | |
| Sparse Autoencoder Evaluation | DINOv2-B activations | Variance Explained87.4 | 6 | |
| Sparse Autoencoding | DINOv2 ImageNet-1k Base (patch tokens) | MSE0.191 | 4 |
Showing 5 of 5 rows