List your company

Safety Evaluation on Alpaca

0HRRBefore Fine-tuning
-1.6449.45320.5531.647Apr 19, 2026

Evaluation results

MethodLinks
Before Fine-tuningBackbone=Llama-2-chat-7B2026.040——
Ours (I)Backbone=Llama-2-chat-7B, Rank=322026.040——
Ours (cannot)Backbone=Llama-2-chat-7B, Rank=322026.040.6——
Before Fine-tuningBackbone=Llama-3.1-Instruct-8B2026.041.4——
SaLoRABackbone=Llama-3.1-Instruct-8B, Rank=322026.041.4——
Ours (I)Backbone=Llama-3.1-Instruct-8B, Rank=322026.041.5——
Ours (cannot)Backbone=Llama-3.1-Instruct-8B, Rank=322026.041.5——
SaLoRABackbone=Llama-3.1-Instruct-8B, Rank=162026.042.9——
Ours (I)Backbone=Llama-2-chat-7B, Rank=162026.042.9——
Ours (I)Backbone=Llama-3.1-Instruct-8B, Rank=162026.043.1——
Ours (cannot)Backbone=Llama-2-chat-7B, Rank=162026.043.4——
SaLoRABackbone=Llama-2-chat-7B, Rank=162026.043.5——
Ours (cannot)Backbone=Llama-3.1-Instruct-8B, Rank=162026.043.5——
SaLoRABackbone=Llama-2-chat-7B, Rank=322026.044.4——
LORA w. IABackbone=Llama-3.1-Instruct-8B, Rank=322026.045.8——
Safe LoRABackbone=Llama-3.1-Instruct-8B, Rank=322026.046.7——
LORA w. IABackbone=Llama-3.1-Instruct-8B, Rank=162026.047.7——
Safe LoRABackbone=Llama-3.1-Instruct-8B, Rank=162026.048.5——
DORABackbone=Llama-3.1-Instruct-8B, Rank=322026.049.4——
DORABackbone=Llama-3.1-Instruct-8B, Rank=162026.0410.1——
LORA w. IABackbone=Llama-2-chat-7B, Rank=162026.0413.5——
LORABackbone=Llama-3.1-Instruct-8B, Rank=162026.0413.8——
PiSSABackbone=Llama-3.1-Instruct-8B, Rank=162026.0413.8——
LORABackbone=Llama-3.1-Instruct-8B, Rank=322026.0414.5——
PiSSABackbone=Llama-3.1-Instruct-8B, Rank=322026.0414.5——
Safe LoRABackbone=Llama-2-chat-7B, Rank=322026.0414.5——
Safe LoRABackbone=Llama-2-chat-7B, Rank=162026.0415.7——
LORA w. VacBackbone=Llama-2-chat-7B, Rank=162026.0420.2——
LORABackbone=Llama-2-chat-7B, Rank=162026.0423.7——
DORABackbone=Llama-2-chat-7B, Rank=162026.0423.7——
LORA w. IABackbone=Llama-2-chat-7B, Rank=322026.0423.7——
DORABackbone=Llama-2-chat-7B, Rank=322026.0425.3——
LORA w. VacBackbone=Llama-2-chat-7B, Rank=322026.0425.3——
LORABackbone=Llama-2-chat-7B, Rank=322026.0431.7——
PiSSABackbone=Llama-2-chat-7B, Rank=162026.0431.7——
PiSSABackbone=Llama-2-chat-7B, Rank=322026.0435.7——
LORA w. VacBackbone=Llama-3.1-Instruct-8B, Rank=322026.0438.3——
LORA w. VacBackbone=Llama-3.1-Instruct-8B, Rank=162026.0441.1——
Goal PrioritizationTarget Model=Vicuna-7B2026.04—130.1
Llama Guard 3Target Model=Vicuna-7B2026.04—038.5
No DefenseTarget Model=Vicuna-7B2026.04—00
Qwen3GuardTarget Model=Vicuna-7B2026.04—042.8
Self-GuardTarget Model=Vicuna-7B2026.04—7.524.6
TrajGuardTarget Model=Vicuna-7B2026.04—05.2