Share your thoughts, 1 month free Claude Pro on usSee more

Safety Evaluation on ToxiGen

100Safety

VCL

Updated 1mo ago

Evaluation Results

Method	Links
VCL 2026.03		100
Vector Ablation 2026.03		98
System Prompt 2026.03		96
Llama-3.1-8B-SFT 2026.03		94
DRO 2026.03		94
Self-Improving Pretraining 2026.01		93.1
Self-CD 2026.03		93
SFT 2025.12		86.7
Llama Pretrain Baseline 2026.01		82
Llama Base 2026.01		80.1
LLaMA2-7B 2025.12		79.1
Contrastive Decoding 2025.12		73.5
Task Vector 2025.12		73.1
Safe Activation 2025.12		71.9
MERGEPRINT 2024.10		54
Feature Subtraction 2025.12		53.5
TRAP 2024.10		53
MERGEPRINT 2024.10		53
TRAP 2024.10		52
TRAP 2024.10		52
MERGEPRINT 2024.10		52
TRAP 2024.10		51
MERGEPRINT 2024.10		51
TRAP 2024.10		50
TRAP 2024.10		50
TRAP 2024.10		50
TRAP 2024.10		50
IF 2024.10		50
MERGEPRINT 2024.10		50
TRAP 2024.10		49
IF 2024.10		49
IF 2024.10		49
IF 2024.10		49
MERGEPRINT 2024.10		49
MERGEPRINT 2024.10		49
MERGEPRINT 2024.10		49
IF 2024.10		48
IF 2024.10		48
IF 2024.10		48
MERGEPRINT 2024.10		48
TRAP 2024.10		47
MERGEPRINT 2024.10		47
TRAP 2024.10		46
TRAP 2024.10		46
IF 2024.10		46
TRAP 2024.10		45
IF 2024.10		45
IF 2024.10		45
IF 2024.10		45
MERGEPRINT 2024.10		45
MERGEPRINT 2024.10		45
MERGEPRINT 2024.10		45
TRAP 2024.10		44
TRAP 2024.10		44
MERGEPRINT 2024.10		44
MERGEPRINT 2024.10		44
MERGEPRINT 2024.10		44
TRAP 2024.10		43
TRAP 2024.10		43
TRAP 2024.10		43
IF 2024.10		43
IF 2024.10		43
IF 2024.10		43
IF 2024.10		43
IF 2024.10		43
MERGEPRINT 2024.10		43
MERGEPRINT 2024.10		43
MERGEPRINT 2024.10		43
MERGEPRINT 2024.10		43
TRAP 2024.10		42
TRAP 2024.10		42
IF 2024.10		42
IF 2024.10		42
IF 2024.10		42
IF 2024.10		42
Model Surgery 2025.12		39.9
DeVec 2025.12		32.1