Visual Question Answering on VizWiz (val-lite)
54.36AccuracyDelimScaling
Evaluation Results
| Method | Links | |
|---|---|---|
| DelimScalingBackbone=Qwen2.5-VL, Model size=7B, Few-shot setting=4-shot interleaved2026.02 | 54.36 | |
| BaselineBackbone=Qwen2.5-VL, Model size=7B, Few-shot setting=4-shot interleaved2026.02 | 53.7 | |
| DelimScalingBackbone=InternVL3, Model size=8B, Few-shot setting=4-shot interleaved2026.02 | 50.92 | |
| BaselineBackbone=InternVL3, Model size=8B, Few-shot setting=4-shot interleaved2026.02 | 47.04 | |
| DelimScalingBackbone=Qwen2.5-VL, Model size=3B, Few-shot setting=4-shot interleaved2026.02 | 42.88 | |
| BaselineBackbone=Qwen2.5-VL, Model size=3B, Few-shot setting=4-shot interleaved2026.02 | 42.38 |