Visual Question Answering on OKVQA (val-lite)
48.68AccuracyDelimScaling
Evaluation Results
| Method | Links | |
|---|---|---|
| DelimScalingBackbone=InternVL3, Model size=8B, Few-shot setting=4-shot interleaved2026.02 | 48.68 | |
| BaselineBackbone=InternVL3, Model size=8B, Few-shot setting=4-shot interleaved2026.02 | 46.84 | |
| DelimScalingBackbone=Qwen2.5-VL, Model size=7B, Few-shot setting=4-shot interleaved2026.02 | 28.24 | |
| BaselineBackbone=Qwen2.5-VL, Model size=7B, Few-shot setting=4-shot interleaved2026.02 | 27.56 | |
| DelimScalingBackbone=Qwen2.5-VL, Model size=3B, Few-shot setting=4-shot interleaved2026.02 | 20 | |
| BaselineBackbone=Qwen2.5-VL, Model size=3B, Few-shot setting=4-shot interleaved2026.02 | 18.04 |