Visual Question Answering on MLLMU-Bench 5% (forget)
55Accuracy (Classification)Vanilla
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| VanillaBackbone=Qwen3-VL-8B, Samples=-2026.05 | 55 | 60 | |
| ASRUBackbone=Qwen3-VL-8B, Samples=27.30%2026.05 | 53.74 | 54 | |
| KL_MinBackbone=Qwen3-VL-8B, Samples=100%2026.05 | 51.67 | 57 | |
| VanillaBackbone=LLaVA-1.5-7B, Samples=-2026.05 | 51.67 | 58 | |
| VanillaBackbone=Qwen3-VL-8B, Samples=-2026.05 | 51.59 | 57 | |
| KL_MinBackbone=Qwen3-VL-8B, Samples=100%2026.05 | 50.66 | 47.5 | |
| GABackbone=Qwen3-VL-8B, Samples=0%2026.05 | 49.17 | 54.9 | |
| GA_diffBackbone=Qwen3-VL-8B, Samples=100%2026.05 | 49.17 | 53.7 | |
| KL_MinBackbone=LLaVA-1.5-7B, Samples=100%2026.05 | 49.17 | 58 | |
| KL_MinBackbone=LLaVA-1.5-7B, Samples=100%2026.05 | 49.09 | 42.6 | |
| MMunlearnerBackbone=LLaVA-1.5-7B, Samples=100%2026.05 | 47.74 | 43.9 | |
| GA_diffBackbone=LLaVA-1.5-7B, Samples=100%2026.05 | 47.5 | 38.8 | |
| GA_diffBackbone=Qwen3-VL-8B, Samples=100%2026.05 | 47.4 | 47.7 | |
| MMunlearnerBackbone=Qwen3-VL-8B, Samples=100%2026.05 | 46.79 | 49.4 | |
| NPOBackbone=Qwen3-VL-8B, Samples=0%2026.05 | 46.72 | 48.9 | |
| VanillaBackbone=Qwen3-VL-4B, Samples=-2026.05 | 46.67 | 57.8 | |
| NPOBackbone=Qwen3-VL-8B, Samples=0%2026.05 | 46.67 | 53.8 | |
| VanillaBackbone=LLaVA-1.5-7B, Samples=-2026.05 | 46.34 | 49 | |
| GA_diffBackbone=Qwen3-VL-4B, Samples=100%2026.05 | 45.83 | 54.8 | |
| GABackbone=Qwen3-VL-8B, Samples=0%2026.05 | 44.81 | 46.9 | |
| KL_MinBackbone=Qwen3-VL-4B, Samples=100%2026.05 | 44.8 | 54.9 | |
| NPOBackbone=LLaVA-1.5-7B, Samples=0%2026.05 | 44.78 | 53.3 | |
| NPOBackbone=LLaVA-1.5-7B, Samples=0%2026.05 | 44.78 | 43.2 | |
| MMunlearnerBackbone=Qwen3-VL-8B, Samples=100%2026.05 | 43.2 | 57.6 | |
| GA_diffBackbone=LLaVA-1.5-7B, Samples=100%2026.05 | 43.17 | 35.1 | |
| VanillaBackbone=Qwen3-VL-4B, Samples=-2026.05 | 41.95 | 54.1 | |
| GABackbone=LLaVA-1.5-7B, Samples=0%2026.05 | 41.67 | 47.4 | |
| ASRUBackbone=LLaVA-1.5-7B, Samples=27.30%2026.05 | 41.35 | 45.4 | |
| GABackbone=LLaVA-1.5-7B, Samples=0%2026.05 | 41.26 | 42.2 | |
| GABackbone=Qwen3-VL-4B, Samples=0%2026.05 | 40.83 | 51.1 | |
| NPOBackbone=Qwen3-VL-4B, Samples=0%2026.05 | 40 | 56.1 | |
| MMunlearnerBackbone=Qwen3-VL-4B, Samples=100%2026.05 | 40 | 55.2 | |
| ASRUBackbone=Qwen3-VL-4B, Samples=27.30%2026.05 | 39.45 | 53.1 | |
| GABackbone=Qwen3-VL-4B, Samples=0%2026.05 | 39.41 | 51.7 | |
| MMunlearnerBackbone=Qwen3-VL-4B, Samples=100%2026.05 | 39.28 | 52.7 | |
| KL_MinBackbone=Qwen3-VL-4B, Samples=100%2026.05 | 38.48 | 51.8 | |
| MMunlearnerBackbone=LLaVA-1.5-7B, Samples=100%2026.05 | 38.33 | 41.5 | |
| GA_diffBackbone=Qwen3-VL-4B, Samples=100%2026.05 | 34.5 | 51.2 | |
| ASRUBackbone=LLaVA-1.5-7B, Samples=27.30%2026.05 | 34.17 | 25 | |
| NPOBackbone=Qwen3-VL-4B, Samples=0%2026.05 | 32.56 | 52 | |
| ASRUBackbone=Qwen3-VL-4B, Samples=27.30%2026.05 | 31.2 | 39.9 | |
| ASRUBackbone=Qwen3-VL-8B, Samples=27.30%2026.05 | 30.83 | 39.8 |