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

Safety Evaluation on MM-SafetyBench v1.0 (test)

0.6ASR

LLaVA-1.5-7B + Mixed-SFT

Updated 1mo ago

Evaluation Results

Method	Links
LLaVA-1.5-7B + Mixed-SFT 2025.03		0.6
LLaVA-1.5-7B + Posthoc-SFT 2025.03		0.6
LLaVA-1.5-7B-LoRA + Mixed-SFT 2025.03		0.6
LLaVA-1.5-7B-LoRA + Posthoc-SFT 2025.03		0.6
LLaVA-1.5-7B + RMU-Unlearning 2025.03		2.98
LLaVA-1.5-7B-LoRA + NPO-Unlearning 2025.03		4.17
LLaVA-1.5-7B + NPO-Unlearning 2025.03		4.76
LLaVA-1.5-7B-LoRA + RMU-Unlearning 2025.03		5.36
LLaVA-1.5-7B + RMU-Unlearning 2025.03		17.26
LLaVA-1.5-7B-LoRA + RMU-Unlearning 2025.03		19.05
LLaVA-1.5-7B + NPO-Unlearning 2025.03		20.24
LLaVA-1.5-7B-LoRA + NPO-Unlearning 2025.03		23.81
LLaVA-1.5-7B + Posthoc-SFT 2025.03		40.48
LLaVA-1.5-7B-LoRA + Posthoc-SFT 2025.03		41.07
LLaVA-1.5-7B 2025.03		48.81
LLaVA-1.5-7B + Mixed-SFT 2025.03		48.81
LLaVA-1.5-7B + Unsafe-Filter 2025.03		50.6
LLaVA-1.5-7B-LoRA 2025.03		57.74
LLaVA-1.5-7B-LoRA + Unsafe-Filter 2025.03		58.93
LLaVA-1.5-7B-LoRA + Mixed-SFT 2025.03		63.69
LLaVA-1.5-7B + Unsafe-Filter 2025.03		90.28
LLaVA-1.5-7B-LoRA + Unsafe-Filter 2025.03		91.07
LLaVA-1.5-7B 2025.03		91.27
LLaVA-1.5-7B-LoRA 2025.03		93.45