SAEV
Collection
SAEs for vision models like CLIP or DINOv2 • 7 items • Updated • 5
SAE for Meta's DINOv3 TopK ViT-L/16 trained on ImageNet-1K Activations
- Homepage: https://osu-nlp-group.github.io/saev
- Code: https://github.com/OSU-NLP-Group/saev
- Preprint: https://arxiv.org/abs/2511.17735
- Demos: https://osu-nlp-group.github.io/saev#demos
- Point of Contact: Sam Stevens
Checkpoints
Each checkpoint is a sparse autoencoder (SAE) trained on a different layer with a different sparsity level. Pick the checkpoint that matches your target layer and desired sparsity (L0).
| Run ID | Layer | L0 | MSE | Path |
|---|---|---|---|---|
| 3ld8ilmo | 13 | 16.0 | 4.0771 | layer_13/3ld8ilmo/sae.pt |
| l03epvhu | 13 | 64.0 | 3.7029 | layer_13/l03epvhu/sae.pt |
| 1up044nl | 13 | 256.0 | 3.5362 | layer_13/1up044nl/sae.pt |
| 6r92o6t6 | 15 | 16.0 | 9.5792 | layer_15/6r92o6t6/sae.pt |
| e4w7u0np | 15 | 64.0 | 8.8184 | layer_15/e4w7u0np/sae.pt |
| 3hzenf5e | 15 | 256.0 | 8.4206 | layer_15/3hzenf5e/sae.pt |
| tkdd41tq | 17 | 16.0 | 22.4154 | layer_17/tkdd41tq/sae.pt |
| jjz6a7ja | 17 | 64.0 | 20.8220 | layer_17/jjz6a7ja/sae.pt |
| huzxe3hu | 17 | 256.0 | 19.6302 | layer_17/huzxe3hu/sae.pt |
| 0c4mlnn7 | 19 | 16.0 | 64.3131 | layer_19/0c4mlnn7/sae.pt |
| 6x4t5t76 | 19 | 64.0 | 60.3351 | layer_19/6x4t5t76/sae.pt |
| xk0a9w3g | 19 | 256.0 | 57.5311 | layer_19/xk0a9w3g/sae.pt |
| rez38zbu | 21 | 16.0 | 185.0552 | layer_21/rez38zbu/sae.pt |
| jxxje744 | 21 | 64.0 | 175.6627 | layer_21/jxxje744/sae.pt |
| 2k6kq9f2 | 21 | 256.0 | 167.8735 | layer_21/2k6kq9f2/sae.pt |
| a95jzikd | 23 | 16.0 | 1429.0161 | layer_23/a95jzikd/sae.pt |
| elwq2g19 | 23 | 64.0 | 1349.6457 | layer_23/elwq2g19/sae.pt |
| l8hooa3r | 23 | 256.0 | 1303.2933 | layer_23/l8hooa3r/sae.pt |
This metadata is also available in manifest.jsonl at the repo root for programmatic access.
Usage
from huggingface_hub import hf_hub_download
import saev.nn
path = hf_hub_download("osunlp/SAE_DINOv3_TopK_ViT-L-16_IN1K", "layer_23/l8hooa3r/sae.pt")
sae = saev.nn.load(path)
Inference Instructions
Follow the instructions here.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support