List your company
ResearchFormerly WizWand

Safety Evaluation on SPA-VL

0.2ASRAutoSteer
-1.6811.0123.736.39Oct 15, 2025
Updated 7mo ago

Evaluation results

MethodLinks
AutoSteerModel=LLaVA-OneVision-7B2025.100.2——
MoRASModel=LLaVA-OneVision-7B2025.101.2——
AutoSteerModel=Qwen-VL-Chat2025.102.3——
MoRASModel=Qwen-VL-Chat2025.102.5——
MoRASBackbone=LLaVA-1.5-13B2025.103.4——
CoCAModel=LLaVA-OneVision-7B2025.104.2——
CoCAModel=Qwen-VL-Chat2025.104.2——
ETAModel=Qwen-VL-Chat2025.104.5——
MoRASModel=InternLM-XComposer-2.52025.104.9——
AutoSteerModel=InternLM-XComposer-2.52025.105.1——
FigStepModel=LLaVA-OneVision-7B2025.105.3——
FigStepModel=Qwen-VL-Chat2025.105.7——
CoCAModel=InternLM-XComposer-2.52025.105.9——
FigStepModel=InternLM-XComposer-2.52025.106.8——
MoRASModel=LLaVA-1.5-7B2025.107.6——
ECSOModel=Qwen-VL-Chat2025.107.6——
AutoSteerModel=LLaVA-1.5-7B2025.108.3——
ASTRAModel=Qwen-VL-Chat2025.108.3——
ETAModel=LLaVA-OneVision-7B2025.108.7——
AutoSteerBackbone=LLaVA-1.5-13B2025.109.1——
CoCABackbone=LLaVA-1.5-13B2025.1010.2——
CoCAModel=LLaVA-1.5-7B2025.1010.9——
ASTRAModel=LLaVA-OneVision-7B2025.1011.3——
ECSOModel=LLaVA-OneVision-7B2025.1012.5——
VanillaModel=Qwen-VL-Chat2025.1012.5——
ETAModel=InternLM-XComposer-2.52025.1014——
ETABackbone=LLaVA-1.5-13B2025.1015.1——
VanillaModel=LLaVA-OneVision-7B2025.1015.1——
ECSOBackbone=LLaVA-1.5-13B2025.1015.5——
ETAModel=LLaVA-1.5-7B2025.1017——
ECSOModel=InternLM-XComposer-2.52025.1019.6——
FigStepBackbone=LLaVA-1.5-13B2025.1021.5——
ASTRAModel=InternLM-XComposer-2.52025.1023——
ECSOModel=LLaVA-1.5-7B2025.1023.4——
VanillaModel=InternLM-XComposer-2.52025.1027.6——
FigStepModel=LLaVA-1.5-7B2025.1032.4——
ASTRABackbone=LLaVA-1.5-13B2025.1035.1——
VanillaBackbone=LLaVA-1.5-13B2025.1040.8——
ASTRAModel=LLaVA-1.5-7B2025.1041.5——
VanillaModel=LLaVA-1.5-7B2025.1047.2——
LLaVA-1.5-13BFine-tuning Strategy=Full fine-tuning2025.03—50.1954.47
LLaVA-1.5-13BFine-tuning Strategy=LoRA fine-tuning2025.03—45.6655.22
LLaVA-1.5-7BFine-tuning Strategy=Full fine-tuning2025.03—46.4252.08
LLaVA-1.5-7BFine-tuning Strategy=LoRA fine-tuning2025.03—44.9150.44
Mixed-SFTBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=Full fine-tuning2025.03—14.3437.73
Mixed-SFTBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=LoRA fine-tuning2025.03—21.5140.13
Mixed-SFTBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=Full fine-tuning2025.03—18.1140.5
Mixed-SFTBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=LoRA fine-tuning2025.03—14.3438.74
NPO-UnlearningBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=Full fine-tuning2025.03—18.4924.15
NPO-UnlearningBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=LoRA fine-tuning2025.03—21.5125.28
NPO-UnlearningBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=Full fine-tuning2025.03—22.2626.04
NPO-UnlearningBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=LoRA fine-tuning2025.03—18.1122.26
Posthoc-SFTBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=Full fine-tuning2025.03—13.5832.96
Posthoc-SFTBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=LoRA fine-tuning2025.03—20.3837.61
Posthoc-SFTBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=Full fine-tuning2025.03—16.9832.83
Posthoc-SFTBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=LoRA fine-tuning2025.03—12.0830.44
RMU-UnlearningBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=Full fine-tuning2025.03—17.7322.64
RMU-UnlearningBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=LoRA fine-tuning2025.03—20.3824.24
RMU-UnlearningBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=Full fine-tuning2025.03—2023.77
RMU-UnlearningBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=LoRA fine-tuning2025.03—17.7422.64
Unsafe-FilterBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=Full fine-tuning2025.03—45.6654.72
Unsafe-FilterBase Model=LLaVA-1.5-7B, Fine-tuning Strategy=LoRA fine-tuning2025.03—45.2852.33
Unsafe-FilterBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=Full fine-tuning2025.03—52.0856.27
Unsafe-FilterBase Model=LLaVA-1.5-13B, Fine-tuning Strategy=LoRA fine-tuning2025.03—48.355.85