List your company
ResearchFormerly WizWand

Classification on MLLMU-Bench (test)

52.5AccuracyGA_Diff
1.363214.639127.91541.1909Feb 21, 2025May 8, 2025Jul 24, 2025Oct 8, 2025Dec 24, 2025Mar 10, 2026May 26, 2026
Updated 5mo ago

Evaluation results

MethodLinks
GA_DiffBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0552.5
BaseBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0551.67
GABackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0551.67
POBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0551.67
KL_MinBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0550.83
NPOBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0550.83
VanillaBase Model=LLaVA-1.5-7B, Forget Percentage=5%2025.0247.86
VGIDBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0547.5
Noise-TBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0546.67
ICU-TBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0545.83
ICU-TBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0542.5
MANUBase Model=LLaVA-1.5-7B, Forget Percentage=5%2025.0241.67
Noise-TBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0540.83
POBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0540
GABackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0538.33
GA_DiffBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0538.33
ICU-TBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0538.33
BaseBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0537.5
NPOBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0537.5
RMUBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0537.5
KL_MinBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0536.67
POBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0535.83
RMUBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0535.83
VGIDBackbone=LLaVA-1.5-7B, Forget Ratio=5%2026.0535
BaseBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0530.83
NPOBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0529.17
GABackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0525.83
Noise-TBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0525
VGIDBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0525
KL_MinBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.0523.67
RMUBackbone=Qwen2.5-VL-3B, Forget Ratio=5%2026.0520.83
GA_DiffBackbone=LLaVA-1.5-13B, Forget Ratio=5%2026.053.33