Question Answering on MLLMU-Bench Real-world 1.0
78.3AccuracyNPO
Evaluation results
| Method | Links | |
|---|---|---|
| NPOBackbone=Qwen2-VL-7B2025.12 | 78.3 | |
| GABackbone=Qwen2-VL-7B2025.12 | 77.8 | |
| VanillaBackbone=Qwen2-VL-7B2025.12 | 77.5 | |
| Visual Knowledge DistillationBackbone=Qwen2-VL-7B2025.12 | 77.4 | |
| MMUBackbone=Qwen2-VL-7B2025.12 | 77.3 | |
| KL MinBackbone=Qwen2-VL-7B2025.12 | 77.1 | |
| GA DiffBackbone=Qwen2-VL-7B2025.12 | 77 | |
| MANUBackbone=Qwen2-VL-7B2025.12 | 76.2 | |
| GABackbone=LLaVA-1.5-7B2025.12 | 55 | |
| NPOBackbone=LLaVA-1.5-7B2025.12 | 55 | |
| VanillaBackbone=LLaVA-1.5-7B2025.12 | 54.9 | |
| MMUBackbone=LLaVA-1.5-7B2025.12 | 54.9 | |
| Visual Knowledge DistillationBackbone=LLaVA-1.5-7B2025.12 | 54.9 | |
| MANUBackbone=LLaVA-1.5-7B2025.12 | 54.8 | |
| KL MinBackbone=LLaVA-1.5-7B2025.12 | 54.6 | |
| GA DiffBackbone=LLaVA-1.5-7B2025.12 | 53.6 |