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: 4
- 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: 171618304
- Learning Rate: 0.0002
- L1 Coefficient: 0.3000
- Batch Size: 4096
- Context Size: 1
Performance Metrics
Sparsity
- L0 (Active Features): 64
- Dead Features: 27636
- Mean Log10 Feature Sparsity: -9.1191
- Features Below 1e-5: 48474
- Features Below 1e-6: 39986
- Mean Passes Since Fired: 14514.3223
Reconstruction
- Explained Variance: 0.9411
- Explained Variance Std: 0.0165
- MSE Loss: 0.0001
- L1 Loss: 0
- Overall Loss: 0.0001
Training Details
- Training Duration: 17915.6712 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/bqv5jw5n
- Original Checkpoint Path: /network/scratch/s/sonia.joseph/checkpoints/clip-b
- Random Seed: 42
- Downloads last month
- 8