Imagenet Sae Top K 64 Patches Only Layer 7 Hook Resid Post 256 87

# CLIP Sparse Autoencoder Checkpoint This model is a sparse autoencoder trained on CLIP's internal representations. ## Model Details ### Architecture - **Layer**: 7 - **Layer Type**: hook_resid_post - **Model**: open-clip:laion/CLIP-ViT-B-32-DataComp.XL-s13B-b90K - **Dictionary Size**: 49152 - **Input Dimension**: 768 - **Expansion Factor**: 64 - **CLS Token Only**: False ### Training - **Training Images**: 1299936 - **Learning Rate**: 0.0002 - **L1 Coefficient**: 0.0002 - **Batch Size**: 4096 - **Context Size**: 49 ## Performance Metrics ### Sparsity - **L0 (Active Features)**: 256.0000 - **Dead Features**: 0 - **Mean Log10 Feature Sparsity**: -4.8271 - **Features Below 1e-5**: 21240 - **Features Below 1e-6**: 1 - **Mean Passes Since Fired**: 24.7823 ### Reconstruction - **Explained Variance**: 0.8733 - **Explained Variance Std**: 0.0286 - **MSE Loss**: 0.0015 - **L1 Loss**: 0 - **Overall Loss**: 0.0015 ## Training Details - **Training Duration**: 2915 seconds - **Final Learning Rate**: 0.0000 - **Warm Up Steps**: 500 - **Gradient Clipping**: 1 ## Additional Information - **Original Checkpoint Path**: /network/scratch/p/praneet.suresh/celeba_checkpoints/2b907d27-tinyclip_sae_16_hyperparam_sweep_lr/n_images_1300020.pt - **Wandb Run**: https://wandb.ai/perceptual-alignment/imagenet-sweep-topk-patches_all_layers/runs/0o6l7k5w - **Random Seed**: 42

Source context: 5 downloads · 0 likes · Repo Prisma-Multimodal/imagenet-sae-top_k-64-patches_only-layer_7-hook_resid_post-256-87