enrich README with config, loading, citation
Browse files
README.md
CHANGED
|
@@ -1,3 +1,24 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# SONAR SAEs — additional runs
|
| 2 |
|
| 3 |
-
Additional SAE training runs on SONAR embeddings
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: cc-by-4.0
|
| 3 |
+
tags:
|
| 4 |
+
- sparse-autoencoder
|
| 5 |
+
- sae
|
| 6 |
+
- sonar
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
# SONAR SAEs — additional runs
|
| 10 |
|
| 11 |
+
Additional SAE training runs on SONAR embeddings that are not directly
|
| 12 |
+
cited in:
|
| 13 |
+
|
| 14 |
+
> **Interpretability of Text Auto-Encoders using Sparse Auto-Encoders: A Sandbox for Interpreting Neuralese.**
|
| 15 |
+
> Nicky Pochinkov & Jason Rich Darmawan, EACL 2026 (submitted).
|
| 16 |
+
|
| 17 |
+
Provided for completeness and reproducibility (early
|
| 18 |
+
hyper-parameter sweeps, exploratory runs that didn't make the paper).
|
| 19 |
+
Match wandb run IDs against
|
| 20 |
+
[`nickypro/sonar-saes-wandb-logs`](https://huggingface.co/nickypro/sonar-saes-wandb-logs) for training metrics.
|
| 21 |
+
|
| 22 |
+
The canonical SAEs are in
|
| 23 |
+
[`nickypro/sonar-saes-large`](https://huggingface.co/nickypro/sonar-saes-large) (scaled-up BatchTopK) and
|
| 24 |
+
[`nickypro/sonar-saes-comparison`](https://huggingface.co/nickypro/sonar-saes-comparison) (four-variant comparison).
|