YAML Metadata
Warning:
empty or missing yaml metadata in repo card
(https://huggingface.co/docs/hub/model-cards#model-card-metadata)
CLIP Sparse Autoencoder Checkpoint
This model is a sparse autoencoder trained on CLIP's internal representations.
Model Details
Architecture
- Layer: 10
- 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: True
Training
- Training Images: 114683904
- Learning Rate: 0.0002
- L1 Coefficient: 0.3000
- Batch Size: 4096
- Context Size: 1
Performance Metrics
Sparsity
- L0 (Active Features): 64
- Dead Features: 0
- Mean Log10 Feature Sparsity: -3.5579
- Features Below 1e-5: 83
- Features Below 1e-6: 2
- Mean Passes Since Fired: 22.8832
Reconstruction
- Explained Variance: 0.8698
- Explained Variance Std: 0.0456
- MSE Loss: 0.0016
- L1 Loss: 0
- Overall Loss: 0.0016
Training Details
- Training Duration: 17901.0753 seconds
- Final Learning Rate: 0.0002
- Warm Up Steps: 200
- Gradient Clipping: 1
Additional Information
- Weights & Biases Run: https://wandb.ai/perceptual-alignment/clip/runs/10hacih1
- Original Checkpoint Path: /network/scratch/s/sonia.joseph/checkpoints/clip-b
- Random Seed: 42
- Downloads last month
- 17