Multimodal Understanding on MMBench EN v1.1 (dev)
76.63AccuracyNGM
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| NGMModel=Qwen3-VL-2B2026.05 | 76.63 | — | — | — | — | — | |
| Qwen3-VL-2BModel=Qwen3-VL-2B2026.05 | 76.32 | — | — | — | — | — | |
| NativeRetained Tokens=5762026.03 | 61.8 | 1 | 1 | 1 | 76.5 | 100 | |
| CDPrunerRetained Tokens=1282026.03 | 60.6 | 1.41 | 0.77 | 0.91 | 73.9 | 96.6 | |
| EvoPruneRetained Tokens=1282026.03 | 59.7 | 1.1 | 0.77 | 0.84 | 74.9 | 97.9 | |
| DivPruneRetained Tokens=1282026.03 | 59.3 | 1.16 | 0.77 | 0.85 | 72.9 | 95.3 | |
| CDPrunerRetained Tokens=642026.03 | 59.1 | 1.38 | 0.73 | 0.86 | 72.2 | 94.4 | |
| VisPrunerRetained Tokens=1282026.03 | 58.8 | 1.32 | 0.77 | 0.9 | 73.2 | 95.7 | |
| FasterVLMRetained Tokens=1282026.03 | 58.3 | 1.15 | 0.77 | 0.85 | 72.3 | 94.6 | |
| EvoPruneRetained Tokens=642026.03 | 58.3 | 1.07 | 0.73 | 0.77 | 72.9 | 95.3 | |
| DivPruneRetained Tokens=642026.03 | 58.1 | 1.12 | 0.73 | 0.79 | 70.4 | 92 | |
| VisPrunerRetained Tokens=642026.03 | 57.7 | 1.35 | 0.73 | 0.86 | 69.9 | 91.3 | |
| DARTRetained Tokens=1282026.03 | 57.5 | 1.2 | 0.77 | 0.88 | 70.9 | 92.7 | |
| CDPrunerRetained Tokens=322026.03 | 57.4 | 1.22 | 0.71 | 0.8 | 69.6 | 91 | |
| EvoPruneRetained Tokens=322026.03 | 56.8 | 1.06 | 0.71 | 0.75 | 68.7 | 89.8 | |
| FasterVLMRetained Tokens=642026.03 | 56.7 | 1.09 | 0.73 | 0.79 | 68.2 | 89.2 | |
| DivPruneRetained Tokens=322026.03 | 55.7 | 1.1 | 0.71 | 0.77 | 66.8 | 87.4 | |
| VisPrunerRetained Tokens=322026.03 | 53.9 | 1.35 | 0.71 | 0.84 | 65.2 | 85.2 | |
| FasterVLMRetained Tokens=322026.03 | 53.7 | 1.09 | 0.71 | 0.77 | 64.5 | 84.4 | |
| DARTRetained Tokens=642026.03 | 53.1 | 1.13 | 0.73 | 0.8 | 66.2 | 86.6 | |
| DARTRetained Tokens=322026.03 | 49.1 | 1.09 | 0.71 | 0.76 | 59.8 | 78.1 |