Update README.md
Browse files
README.md
CHANGED
|
@@ -24,8 +24,6 @@ Built on Qwen3-0.6B. Trained with **mechanical supervision** (NVIDIA Aegis-2.0 l
|
|
| 24 |
| Aegis-2.0 test F1 (in-domain detection) | **76%** (P69/R85) | NVIDIA's primary in-domain metric; in the range of 8B guards |
|
| 25 |
| ToxicChat F1 (out-of-domain) | 22% (P14/R55) | high recall, over-flags novel benign content; see limitations |
|
| 26 |
|
| 27 |
-
Incumbent sub-1B guards score ~0.7% on policy-flip.
|
| 28 |
-
|
| 29 |
## Railz vs Railz-R (pick one)
|
| 30 |
| | flip | Aegis-F1 | ToxicChat | use when |
|
| 31 |
|---|---|---|---|---|
|
|
|
|
| 24 |
| Aegis-2.0 test F1 (in-domain detection) | **76%** (P69/R85) | NVIDIA's primary in-domain metric; in the range of 8B guards |
|
| 25 |
| ToxicChat F1 (out-of-domain) | 22% (P14/R55) | high recall, over-flags novel benign content; see limitations |
|
| 26 |
|
|
|
|
|
|
|
| 27 |
## Railz vs Railz-R (pick one)
|
| 28 |
| | flip | Aegis-F1 | ToxicChat | use when |
|
| 29 |
|---|---|---|---|---|
|