Visual Question Answering on CLEAR Forget 1.0
34.2AccuracyVisual Knowledge Distillation
Evaluation results
| Method | Links | |
|---|---|---|
| Visual Knowledge DistillationBackbone=LLaVA-1.5-7B2025.12 | 34.2 | |
| MMUBackbone=LLaVA-1.5-7B2025.12 | 36.2 | |
| MANUBackbone=LLaVA-1.5-7B2025.12 | 39.8 | |
| KL MinBackbone=LLaVA-1.5-7B2025.12 | 40.4 | |
| NPOBackbone=LLaVA-1.5-7B2025.12 | 40.4 | |
| GA DiffBackbone=LLaVA-1.5-7B2025.12 | 47.3 | |
| Visual Knowledge DistillationBackbone=Qwen2-VL-7B2025.12 | 48.7 | |
| MMUBackbone=Qwen2-VL-7B2025.12 | 50 | |
| GABackbone=Qwen2-VL-7B2025.12 | 55.3 | |
| GABackbone=LLaVA-1.5-7B2025.12 | 57.4 | |
| MANUBackbone=Qwen2-VL-7B2025.12 | 61.3 | |
| NPOBackbone=Qwen2-VL-7B2025.12 | 62.8 | |
| VanillaBackbone=LLaVA-1.5-7B2025.12 | 63.3 | |
| GA DiffBackbone=Qwen2-VL-7B2025.12 | 63.3 | |
| VanillaBackbone=Qwen2-VL-7B2025.12 | 67 | |
| KL MinBackbone=Qwen2-VL-7B2025.12 | 67 |