|
--- |
|
license: apache-2.0 |
|
--- |
|
|
|
# 1. GemmaScope |
|
|
|
Gemmascope is TODO |
|
|
|
# 2. What Is `gemmascope-2b-pt-att`? |
|
|
|
- `gemmascope-`: See 1. |
|
- `2b-pt-`: These SAEs were trained on Gemma v2 2B base model (TODO link) |
|
- `att`: These SAEs were trained on the attention layer outputs, before the final linear projection (TODO link ckkissane post). |
|
|
|
## 3. GTM FAQ (TODO(conmy): delete for main rollout) |
|
|
|
Q1: Why does this model exist in `gg-hf`? |
|
|
|
A1: See https://docs.google.com/document/d/1bKaOw2mJPJDYhgFQGGVOyBB3M4Bm_Q3PMrfQeqeYi0M (Google internal only). |
|
|
|
Q2: What does "SAE" mean? |
|
|
|
A2: Sparse Autoencoder. See https://docs.google.com/document/d/1roMgCPMPEQgaNbCu15CGo966xRLToulCBQUVKVGvcfM (should be available to trusted HuggingFace collaborators, and Google too). |
|
|
|
TODO(conmy): remove this when making the main repo. |
|
|
|
## 4. Point of Contact |
|
|
|
Point of contact: Arthur Conmy |
|
|
|
Contact by email: |
|
|
|
```python |
|
''.join(list('moc.elgoog@ymnoc')[::-1]) |
|
``` |
|
|
|
HuggingFace account: |
|
https://huggingface.co/ArthurConmyGDM |
|
|
