Safety Alignment on Alpaca-65B Weak-to-strong Safety Alignment (test)
131.81HV ScoreUniARM
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| UniARMGuide model=Alpaca-7B, Frozen target model=Alpaca-65B, Setup=Weak-to-strong extension, Training mode=fine-tuned on Alpaca-7B2026.02 | 131.81 | 3.76 | |
| PARMGuide model=Alpaca-7B, Frozen target model=Alpaca-65B, Setup=Weak-to-strong extension, Training mode=fine-tuned on Alpaca-7B2026.02 | 120.85 | 3.52 | |
| GenARMGuide model=Alpaca-7B, Frozen target model=Alpaca-65B, Setup=Weak-to-strong extension, Training mode=fine-tuned on Alpaca-7B2026.02 | 111.72 | 1.99 |