Multimodal Evaluation on MMB
85.31ScoreBaseline
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Baseline2026.02 | 85.31 | — | — | — | — | — | — | |
| IDPrunerRetain Tokens=25%2026.02 | 81.27 | — | — | — | — | — | — | |
| DivPruneRetain Tokens=25%2026.02 | 80.67 | — | — | — | — | — | — | |
| VisionSelectorRetain Tokens=25%2026.02 | 80.07 | — | — | — | — | — | — | |
| DARTRetain Tokens=25%2026.02 | 79.81 | — | — | — | — | — | — | |
| SCOPERetain Tokens=25%2026.02 | 78.18 | — | — | — | — | — | — | |
| FastVRetain Tokens=25%2026.02 | 77.66 | — | — | — | — | — | — | |
| VisPrunerRetain Tokens=25%2026.02 | 76.55 | — | — | — | — | — | — | |
| IDPrunerRetain Tokens=10%2026.02 | 75.69 | — | — | — | — | — | — | |
| VisionZipRetain Tokens=25%2026.02 | 75.26 | — | — | — | — | — | — | |
| VisionSelectorRetain Tokens=10%2026.02 | 74.91 | — | — | — | — | — | — | |
| DivPruneRetain Tokens=10%2026.02 | 73.8 | — | — | — | — | — | — | |
| SCOPERetain Tokens=10%2026.02 | 73.54 | — | — | — | — | — | — | |
| HiPruneRetain Tokens=25%2026.02 | 72.51 | — | — | — | — | — | — | |
| DARTRetain Tokens=10%2026.02 | 70.19 | — | — | — | — | — | — | |
| FastVRetain Tokens=10%2026.02 | 69.07 | — | — | — | — | — | — | |
| TinyLLaVALLM=2.7B, Res.=3362024.06 | 68.3 | — | — | — | — | — | — | |
| VisPrunerRetain Tokens=10%2026.02 | 67.7 | — | — | — | — | — | — | |
| V2DropModel=LLaVA-NeXT, TFLOPs=70%2026.06 | 67.25 | — | — | — | — | — | 98.4 | |
| APETModel=LLaVA-NeXT, TFLOPs=70%2026.06 | 67.18 | — | — | — | — | — | 99.4 | |
| VanillaModel=LLaVA-NeXT, TFLOPs=100%2026.06 | 67.1 | — | — | — | — | — | 100 | |
| VSkip+ (N=20)Model=LLaVA-NeXT, TFLOPs=74%2026.06 | 67.01 | — | — | — | — | — | 98.2 | |
| ShortV (N=20)Model=LLaVA-NeXT, TFLOPs=51%2026.06 | 66.92 | — | — | — | — | — | 96 | |
| SPHINXLLM=7B, Res.=2242024.06 | 66.9 | — | — | — | — | — | — | |
| VTW (K=16)Model=LLaVA-NeXT, TFLOPs=51%2026.06 | 66.75 | — | — | — | — | — | 86.5 | |
| Ours (N=20)Model=LLaVA-NeXT, TFLOPs=67%2026.06 | 66.75 | — | — | — | — | — | 98.8 | |
| VisionZipRetain Tokens=10%2026.02 | 65.38 | — | — | — | — | — | — | |
| LLaVA1.5LLM=7B, Res.=3362024.06 | 64.3 | — | — | — | — | — | — | |
| HiPruneRetain Tokens=10%2026.02 | 63.83 | — | — | — | — | — | — | |
| RoboMambaLLM=2.7B, Res.=2242024.06 | 60.9 | — | — | — | — | — | — | |
| RoboMambaLLM=2.7B, Res.=3362024.06 | 60.7 | — | — | — | — | — | — | |
| LLaVA-PhiLLM=2.7B, Res.=3362024.06 | 59.8 | — | — | — | — | — | — | |
| MobileVLMLLM=2.7B, Res.=3362024.06 | 59.6 | — | — | — | — | — | — | |
| InstructBLIPLLM=7B, Res.=2242024.06 | 36 | — | — | — | — | — | — | |
| FlashSlothParameters=3.2B2024.12 | — | 90 | 0.09 | 7.6 | 0.05 | 11.7 | — | |
| FlashSloth-HDParameters=3.2B2024.12 | — | 414 | 1.98 | 8.7 | 0.11 | 5.1 | — | |
| IMPParameters=3.1B2024.12 | — | 729 | 0.31 | 8.4 | 0.12 | 8.4 | — | |
| InternVL2Parameters=2B2024.12 | — | 1,296 | 3.98 | 11.5 | 0.19 | 4.7 | — | |
| LLaVAParameters=7B, Version=V1.52024.12 | — | 576 | 4.77 | 16.4 | 0.11 | 6.1 | — | |
| Qwen2-VLParameters=2B2024.12 | — | 385 | 0.84 | 8 | 0.09 | 7 | — |