Prisma-Multimodal/DINO-vanilla-x64-all_patches_9-resid_post-1105-99

Model

0

stars

4

commits

5

linked in READMEs

Apr 17, 2025

updated

Browse cluster: Self-Supervised Vision Model Training →

README

DINO Sparse Autoencoder Checkpoint

This model is a sparse autoencoder trained on DINO's internal representations.

Model Details

Architecture

Layer: 9
Layer Type: hook_resid_post
Model: facebook/dino-vitb16
Dictionary Size: 49152
Input Dimension: 768
Expansion Factor: 64
CLS Token Only: False

Training

Learning Rate: 4.4606619925618766e-05
Batch Size: 4096
Context Size: 197

Sparsity

L0 (Active Features): 1105

Additional Information

Wandb Run: https://wandb.ai/perceptual-alignment/DINO-SAE-Vanilla-Sweep/runs/5bsk6gj1
Random Seed: 42

Contributors

4 commits

Similar repos

Prisma-Multimodal/DINO-vanilla-x64-all_patches_11-resid_post-1085-99

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_1-resid_post-549-95

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_0-resid_post-507-98

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_5-resid_post-1001-98

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_3-resid_post-989-95

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_10-resid_post-835-99

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_6-resid_post-962-99

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_4-resid_post-876-99

0

Linked in READMEs

Prisma-Multimodal/ViT-Prisma

ViT Prisma is a mechanistic interpretability library for Vision and Video Transformers (ViTs).

388

soniajoseph/ViT-Prisma

ViT Prisma is a mechanistic interpretability library for Vision and Video Transformers (ViTs).

388

leeloolee/df-prisma

0

natihash/my_prisma

0

alexlav17/ViT-Prisma-fix

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_9-resid_post-1105-99

Model

0

stars

4

commits

5

linked in READMEs

Apr 17, 2025

updated

Browse cluster: Self-Supervised Vision Model Training →

README

DINO Sparse Autoencoder Checkpoint

This model is a sparse autoencoder trained on DINO's internal representations.

Model Details

Architecture

Layer: 9
Layer Type: hook_resid_post
Model: facebook/dino-vitb16
Dictionary Size: 49152
Input Dimension: 768
Expansion Factor: 64
CLS Token Only: False

Training

Learning Rate: 4.4606619925618766e-05
Batch Size: 4096
Context Size: 197

Sparsity

L0 (Active Features): 1105

Additional Information

Wandb Run: https://wandb.ai/perceptual-alignment/DINO-SAE-Vanilla-Sweep/runs/5bsk6gj1
Random Seed: 42

Similar repos

Prisma-Multimodal/DINO-vanilla-x64-all_patches_11-resid_post-1085-99

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_1-resid_post-549-95

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_0-resid_post-507-98

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_5-resid_post-1001-98

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_3-resid_post-989-95

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_10-resid_post-835-99

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_6-resid_post-962-99

0

Prisma-Multimodal/DINO-vanilla-x64-all_patches_4-resid_post-876-99

0

Linked in READMEs

Prisma-Multimodal/ViT-Prisma

ViT Prisma is a mechanistic interpretability library for Vision and Video Transformers (ViTs).

388

soniajoseph/ViT-Prisma

ViT Prisma is a mechanistic interpretability library for Vision and Video Transformers (ViTs).

388

leeloolee/df-prisma

0

natihash/my_prisma

0

alexlav17/ViT-Prisma-fix

0

Contributors

4 commits