Commit ·
c1d3443
1
Parent(s): 23dfd84
Update README.md
Browse files
README.md
CHANGED
|
@@ -24,7 +24,6 @@ The goal is to create a model that does not judge or label its users, while main
|
|
| 24 |
|
| 25 |
## What's Different
|
| 26 |
|
| 27 |
-
The refusal circuits were surgically targeted across **layers 19–46**
|
| 28 |
- Refusals dropped from **91/100 to 6/100**
|
| 29 |
- KL divergence : **0.0678**
|
| 30 |
|
|
|
|
| 24 |
|
| 25 |
## What's Different
|
| 26 |
|
|
|
|
| 27 |
- Refusals dropped from **91/100 to 6/100**
|
| 28 |
- KL divergence : **0.0678**
|
| 29 |
|