Multimodal Benchmarking on MMBench (MMB)
65.4MMB ScoreHoloV
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HoloVBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/ VE, Venue=NeurIPS’252026.02 | 65.4 | 98.7 | |
| HoloVLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 65.4 | — | |
| VanillaBackbone=LLaVA-1.5-7B, Token Budget=5762026.02 | 64.7 | 100 | |
| FastV+VTWRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 64.7 | — | |
| FastV+RandomRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 64.7 | — | |
| VanillaLoc.=-, Retained Tokens=576, Pruning Ratio=0%2026.03 | 64.7 | — | |
| LLaVA-1.5-7BRetained Tokens=576, Reduction Ratio=0%2025.12 | 64.6 | — | |
| PIO-FVLMBackbone=LLaVA-1.5-7B, Token Budget=192, VE=w/ VE2026.02 | 64.2 | 98.9 | |
| DART+VTWRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 64.1 | — | |
| DART+RandomRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 63.9 | — | |
| HoloVLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 63.9 | — | |
| DARTRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 63.7 | — | |
| PRUNESIDLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 63.7 | — | |
| DARTLoc.=L2, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 63.6 | — | |
| PDropLoc.=L8–L24, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 63.6 | — | |
| SparseVLMRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 63.4 | — | |
| AutoSelectLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 63.4 | — | |
| PDropRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 63.3 | — | |
| HoloVLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 63.3 | — | |
| FastVRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 63.2 | — | |
| DivPrune+VTWRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 63.2 | — | |
| DivPrune+RandomRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 63.2 | — | |
| DARTLoc.=L2, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 63.2 | — | |
| VisionZipLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 63 | — | |
| AutoSelectLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 62.9 | — | |
| HiREDLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 62.8 | — | |
| AutoSelectLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 62.6 | — | |
| SparseVLMLoc.=L1–L32, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 62.5 | — | |
| DivPruneRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 62.3 | — | |
| DivPruneLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 62.3 | — | |
| DivPruneLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 62.3 | — | |
| PRUNESIDLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 62.1 | — | |
| VisionZipLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 62 | — | |
| DivPrune+VTWRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 61.6 | — | |
| PDropLoc.=L8–L24, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 61.6 | — | |
| HiREDLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 61.5 | — | |
| DivPrune+RandomRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 61.3 | — | |
| VTWRetained Tokens=192, Reduction Ratio=66.7%2025.12 | 61.2 | — | |
| FastVLoc.=L2–L32, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 61.2 | — | |
| DARTLoc.=L2, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 60.6 | — | |
| ToMeLoc.=Pre, Retained Tokens=192, Pruning Ratio=66.7%2026.03 | 60.5 | — | |
| DART+RandomRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 60.4 | — | |
| DART+VTWRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 60.3 | — | |
| HiREDLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 60.2 | — | |
| VisionZipLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 60.1 | — | |
| SparseVLMLoc.=L1–L32, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 60 | — | |
| DivPruneLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 59.3 | — | |
| PDropLoc.=L8–L24, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 58.8 | — | |
| PRUNESIDLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 58.8 | — | |
| SparseVLMRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 58.6 | — | |
| SparseVLMLoc.=L1–L32, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 56.2 | — | |
| FastVLoc.=L2–L32, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 56.1 | — | |
| DARTRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 55.8 | — | |
| FastV+RandomRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 54.9 | — | |
| DivPruneRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 54.6 | — | |
| FastV+VTWRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 54.6 | — | |
| ToMeLoc.=Pre, Retained Tokens=128, Pruning Ratio=77.8%2026.03 | 53.3 | — | |
| FastVLoc.=L2–L32, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 48 | — | |
| PDropRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 45.9 | — | |
| ToMeLoc.=Pre, Retained Tokens=64, Pruning Ratio=88.9%2026.03 | 43.7 | — | |
| FastVRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 34.1 | — | |
| VTWRetained Tokens=64, Reduction Ratio=88.9%2025.12 | 22 | — |