| license: cc-by-4.0 | |
| tags: | |
| - sparse-autoencoder | |
| - sae | |
| - sonar | |
| # SONAR SAEs — additional runs | |
| Additional SAE training runs on SONAR embeddings that are not directly | |
| cited in: | |
| > **Interpretability of Text Auto-Encoders using Sparse Auto-Encoders: A Sandbox for Interpreting Neuralese.** | |
| > Nicky Pochinkov & Jason Rich Darmawan, EACL 2026 (submitted). | |
| Provided for completeness and reproducibility (early | |
| hyper-parameter sweeps, exploratory runs that didn't make the paper). | |
| Match wandb run IDs against | |
| [`nickypro/sonar-saes-wandb-logs`](https://huggingface.co/nickypro/sonar-saes-wandb-logs) for training metrics. | |
| The canonical SAEs are in | |
| [`nickypro/sonar-saes-large`](https://huggingface.co/nickypro/sonar-saes-large) (scaled-up BatchTopK) and | |
| [`nickypro/sonar-saes-comparison`](https://huggingface.co/nickypro/sonar-saes-comparison) (four-variant comparison). | |