Safety Alignment on HEx-PHI
98.8HEx-PHI ScoreDiaBlo
Evaluation Results
| Method | Links | |
|---|---|---|
| DiaBloModel=Mistral-7B, r/N=N = 64, #Params=1.68%2025.06 | 98.8 | |
| DiaBloModel=Mistral-7B, r/N=N = 128, #Params=0.84%2025.06 | 98.8 | |
| DiaBloModel=Llama3-8B, r/N=N = 64, #Params=1.51%2025.06 | 97.6 | |
| HELLoRAParams=0.31%, Backbone=Mixtral-8x7B2026.05 | 96.67 | |
| DiaBloModel=Llama3-8B, r/N=N = 128, #Params=0.76%2025.06 | 96.3 | |
| HELLoRI-SParams=0.02%, Backbone=Mixtral-8x7B2026.05 | 96.06 | |
| DoRAModel=Mistral-7B, r/N=r = 32, #Params=1.25%2025.06 | 95.3 | |
| HELLoRI-DParams=0.17%, Backbone=Mixtral-8x7B2026.05 | 94.85 | |
| LoRIModel=Mistral-7B, r/N=r = 32, #Params=0.63%2025.06 | 94.7 | |
| LoRAParams=1.03%, Backbone=Mixtral-8x7B2026.05 | 93.94 | |
| LoRI-SParams=0.06%, Backbone=Mixtral-8x7B2026.05 | 93.64 | |
| DoRAModel=Llama3-8B, r/N=r = 32, #Params=1.12%2025.06 | 93.6 | |
| LoRI-DParams=0.59%, Backbone=Mixtral-8x7B2026.05 | 93.33 | |
| LoRIModel=Llama3-8B, r/N=r = 32, #Params=0.56%2025.06 | 92.8 | |
| LoRAModel=Mistral-7B, r/N=r = 32, #Params=1.25%2025.06 | 91.9 | |
| LoRAModel=Llama3-8B, r/N=r = 32, #Params=1.12%2025.06 | 91.6 | |
| Zero-shotModel=Mistral-7B, r/N=N/A, #Params=0%2025.06 | 80.5 | |
| Zero-shotModel=Llama3-8B, r/N=N/A, #Params=0%2025.06 | 78.7 |