Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,42 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: apache-2.0
|
| 3 |
-
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: apache-2.0
|
| 3 |
+
---
|
| 4 |
+
|
| 5 |
+
---
|
| 6 |
+
license: apache-2.0
|
| 7 |
+
---
|
| 8 |
+
|
| 9 |
+
# 1. GemmaScope
|
| 10 |
+
|
| 11 |
+
Gemmascope is TODO
|
| 12 |
+
|
| 13 |
+
# 2. What Is `gemmascope-2b-pt-res`?
|
| 14 |
+
|
| 15 |
+
- `gemmascope-`: See <b>1.</b>
|
| 16 |
+
- `2b-pt-`: These SAEs were trained on Gemma v2 2B base model.
|
| 17 |
+
- `res`: These SAEs were trained on the model's residual stream.
|
| 18 |
+
|
| 19 |
+
## 3. GTM FAQ (TODO(conmy): delete for main rollout)
|
| 20 |
+
|
| 21 |
+
Q1: Why does this model exist in `gg-hf`?
|
| 22 |
+
|
| 23 |
+
A1: See https://docs.google.com/document/d/1bKaOw2mJPJDYhgFQGGVOyBB3M4Bm_Q3PMrfQeqeYi0M (Google internal only).
|
| 24 |
+
|
| 25 |
+
Q2: What does "SAE" mean?
|
| 26 |
+
|
| 27 |
+
A2: Sparse Autoencoder. See https://docs.google.com/document/d/1roMgCPMPEQgaNbCu15CGo966xRLToulCBQUVKVGvcfM (should be available to trusted HuggingFace collaborators, and Google too).
|
| 28 |
+
|
| 29 |
+
TODO(conmy): remove this when making the main repo.
|
| 30 |
+
|
| 31 |
+
## 4. Point of Contact
|
| 32 |
+
|
| 33 |
+
Point of contact: Arthur Conmy
|
| 34 |
+
|
| 35 |
+
Contact by email:
|
| 36 |
+
|
| 37 |
+
```python
|
| 38 |
+
''.join(list('moc.elgoog@ymnoc')[::-1])
|
| 39 |
+
```
|
| 40 |
+
|
| 41 |
+
HuggingFace account:
|
| 42 |
+
https://huggingface.co/ArthurConmyGDM
|