Safety alignment on SORRY-Bench (Score, log C)
85.7Scoreπref
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| πrefBackbone=zephyr-7b, N=12026.07 | 85.7 | 5.44 | |
| πrefBackbone=zephyr-7b, N=162026.07 | 73.9 | — | |
| URIALBackbone=zephyr-7b, N=12026.07 | 63 | 3.15 | |
| URIALBackbone=zephyr-7b, N=162026.07 | 52.7 | — | |
| BoBNBackbone=zephyr-7b, N=12026.07 | 43.9 | — | |
| BoBNBackbone=zephyr-7b, N=42026.07 | 41.2 | 2.87 | |
| BoBNBackbone=zephyr-7b, N=162026.07 | 36.1 | — | |
| πrefBackbone=tulu3-8b, N=12026.07 | 28.6 | 1.59 | |
| πrefBackbone=tulu3-8b, N=162026.07 | 26.8 | — | |
| URIALBackbone=tulu3-8b, N=12026.07 | 25.7 | 1.6 | |
| BoBNBackbone=tulu3-8b, N=12026.07 | 24.3 | — | |
| BoBNBackbone=tulu3-8b, N=42026.07 | 23.6 | 1.59 | |
| URIALBackbone=tulu3-8b, N=162026.07 | 22.6 | — | |
| BoBNBackbone=tulu3-8b, N=162026.07 | 21.8 | — |