| # CLIP Transcoder Checkpoint | |
| This model is a Transcoder trained on CLIP's internal representations. | |
| ## Model Details | |
| ### Architecture | |
| - **Layer**: 3 | |
| - **Layer Type**: ln2.hook_normalized | |
| - **Model**: laion/CLIP-ViT-B-32-DataComp.XL-s13B-b90K | |
| - **Dictionary Size**: 49152 | |
| - **Input Dimension**: 768 | |
| - **Expansion Factor**: 64 | |
| - **CLS Token Only**: False | |
| ### Training | |
| - **Learning Rate**: 0.0012988935825045404 | |
| - **Batch Size**: 4096 | |
| - **Context Size**: 50 | |
| ### Sparsity | |
| - **L0 (Active Features)**: 1024 | |
| ## Additional Information | |
| - **Wandb Run**: https://wandb.ai/perceptual-alignment/openclip-transcoders/runs/c9ijpi5h/ | |
| - **Random Seed**: 42 | |