Multimodal Evaluation on MME (score)
2,375.9MME ScoreQwen-2.5-VL-7B + EvoLMM
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen-2.5-VL-7B + EvoLMMsetting=zero-shot2025.11 | 2,375.9 | |
| Qwen2.5-VL-7BToken Budget=100%2025.08 | 2,331 | |
| Qwen-2.5-VL-7Bsetting=zero-shot2025.11 | 2,318.8 | |
| HiPrune++Base Model=Qwen2.5-VL-7B, Token Budget=33.3%2025.08 | 2,310 | |
| HiPruneBase Model=Qwen2.5-VL-7B, Token Budget=33.3%2025.08 | 2,297 | |
| Upper BoundBackbone=InternVL2-26B2026.04 | 2,270 | |
| CLASPBackbone=InternVL2-26B, Token Retention=35%2026.04 | 2,262 | |
| ToMeBackbone=InternVL2-26B, Token Retention=35%2026.04 | 2,178 | |
| HiPrune++Base Model=Qwen2.5-VL-7B, Token Budget=22.2%2025.08 | 2,172 | |
| HiPruneBase Model=Qwen2.5-VL-7B, Token Budget=22.2%2025.08 | 2,168 | |
| Qwen2.5-VL-3BToken Budget=100%2025.08 | 2,144 | |
| FastVBackbone=InternVL2-26B, Token Retention=35%2026.04 | 2,140 | |
| HiPrune++Base Model=Qwen2.5-VL-3B, Token Budget=33.3%2025.08 | 2,069 | |
| HiPruneBase Model=Qwen2.5-VL-3B, Token Budget=33.3%2025.08 | 2,061 | |
| HiPruneBase Model=Qwen2.5-VL-3B, Token Budget=22.2%2025.08 | 2,002 | |
| HiPrune++Base Model=Qwen2.5-VL-7B, Token Budget=11.1%2025.08 | 1,999 | |
| HiPruneBase Model=Qwen2.5-VL-7B, Token Budget=11.1%2025.08 | 1,998 | |
| HiPrune++Base Model=Qwen2.5-VL-3B, Token Budget=22.2%2025.08 | 1,980 | |
| Upper BoundBase Model=LLaVA-NeXT-13B2026.04 | 1,901 | |
| LLaVA-NeXT-13BToken Num=28802025.08 | 1,901 | |
| HiPrune++Backbone=LLaVA-NeXT-13B, Token Num=6402025.08 | 1,894 | |
| HiPruneBase Model=Qwen2.5-VL-3B, Token Budget=11.1%2025.08 | 1,881 | |
| HiPruneBackbone=LLaVA-NeXT-13B, Token Num=6402025.08 | 1,877 | |
| CLASPBase Model=LLaVA-NeXT-13B, Token Retention Budget=640 Tokens2026.04 | 1,867 | |
| UncompressedPruning Mode=/, Pruning Ratio=/, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=false, TFLOPs=5.632026.04 | 1,862 | |
| LLaVA-1.5-7BRetained Tokens=576, Pruning Ratio=100%2026.04 | 1,862 | |
| Upper BoundRetained Tokens=5762026.04 | 1,862 | |
| LLaVA-1.5-7BVTR Type=N/A, Tokens Retained=576, Reduction Ratio=100%2026.06 | 1,862 | |
| DARTPruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=false, TFLOPs=3.182026.04 | 1,856 | |
| Upper BoundBackbone=LLaVA-NeXT 7B, Tokens Retained=28802026.04 | 1,851 | |
| DeSAPRetained Tokens=192, Pruning Ratio=↓ 66.7%2026.04 | 1,848 | |
| CLASPBase Model=LLaVA-NeXT-13B, Token Retention Budget=320 Tokens2026.04 | 1,848 | |
| CLASPRetained Tokens=192, Pruning Ratio=66.7%2026.04 | 1,848 | |
| PDropPruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=true, TFLOPs=3.092026.04 | 1,846 | |
| HiPrune++Base Model=Qwen2.5-VL-3B, Token Budget=11.1%2025.08 | 1,844 | |
| CoMPPruning Mode=C, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=true, TFLOPs=2.942026.04 | 1,843 | |
| FlowCutRetained Tokens=192, Pruning Ratio=↓ 66.7%2026.04 | 1,836 | |
| VisionZipPruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=true, TFLOPs=3.012026.04 | 1,834 | |
| DARTPruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=true, TFLOPs=3.182026.04 | 1,829 | |
| V2DropRetained Tokens=192, Pruning Ratio=↓ 66.7%2026.04 | 1,826 | |
| DARTPruning Mode=T, Pruning Ratio=0.62, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=true, TFLOPs=2.382026.04 | 1,823 | |
| HiPrune++Backbone=LLaVA-NeXT-13B, Token Num=3202025.08 | 1,823 | |
| SparseVLMBase Model=LLaVA-NeXT-13B, Token Retention Budget=640 Tokens2026.04 | 1,821 | |
| FastVVTR Type=Text-aware, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,821 | |
| HoloVRetained Tokens=192, Pruning Ratio=↓ 66.7%2026.04 | 1,820 | |
| LLaVA-1.5-13BToken Num=5762025.08 | 1,818 | |
| VisPruner + RESTOREVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,816 | |
| DeSAPRetained Tokens=128, Pruning Ratio=↓ 77.8%2026.04 | 1,815 | |
| DivPrune + RESTOREVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,813 | |
| VisPruner + RESTOREVTR Type=Text-agnostic, Tokens Retained=128, Reduction Ratio=22.2%2026.06 | 1,813 | |
| HiPrune++Backbone=LLaVA-1.5-13B, Token Num=1922025.08 | 1,808 | |
| HoloV + RESTOREVTR Type=Text-agnostic, Tokens Retained=128, Reduction Ratio=22.2%2026.06 | 1,807 | |
| HoloVRetained Tokens=128, Pruning Ratio=↓ 77.8%2026.04 | 1,802 | |
| CLASPBase Model=LLaVA-NeXT-13B, Token Retention Budget=160 Tokens2026.04 | 1,801 | |
| HiPruneBackbone=LLaVA-NeXT-13B, Token Num=3202025.08 | 1,800 | |
| SparseVLMBase Model=LLaVA-NeXT-13B, Token Retention Budget=320 Tokens2026.04 | 1,798 | |
| HiPruneBackbone=LLaVA-1.5-13B, Token Num=1922025.08 | 1,798 | |
| PDropPruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=false, TFLOPs=3.092026.04 | 1,797 | |
| PyramidDropRetained Tokens=192, Pruning Ratio=↓ 66.7%2026.04 | 1,797 | |
| DivPrune + RESTOREVTR Type=Text-agnostic, Tokens Retained=128, Reduction Ratio=22.2%2026.06 | 1,795 | |
| HoloV + RESTOREVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,793 | |
| FlowCutRetained Tokens=128, Pruning Ratio=↓ 77.8%2026.04 | 1,792 | |
| CLASPRetained Tokens=128, Pruning Ratio=77.8%2026.04 | 1,790 | |
| PDropVTR Type=Text-aware, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,790 | |
| SparseVLMPruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=false, TFLOPs=3.012026.04 | 1,787 | |
| HoloV + RESTOREVTR Type=Text-agnostic, Tokens Retained=64, Reduction Ratio=11.1%2026.06 | 1,787 | |
| VisPrunerVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,784 | |
| VisionZipPruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=false, TFLOPs=3.012026.04 | 1,783 | |
| VisionZipRetained Tokens=192, Pruning Ratio=↓ 66.7%2026.04 | 1,783 | |
| SparseVLMVTR Type=Text-aware, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,782 | |
| VisionZip + RESTOREVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,782 | |
| HoloVVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,779 | |
| VisPrunerVTR Type=Text-agnostic, Tokens Retained=128, Reduction Ratio=22.2%2026.06 | 1,778 | |
| HiPrune++Backbone=LLaVA-NeXT-13B, Token Num=1602025.08 | 1,774 | |
| VisionZip + RESTOREVTR Type=Text-agnostic, Tokens Retained=128, Reduction Ratio=22.2%2026.06 | 1,772 | |
| DARTPruning Mode=T, Pruning Ratio=0.62, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=false, TFLOPs=2.382026.04 | 1,765 | |
| VisionZipRetained Tokens=128, Pruning Ratio=↓ 77.8%2026.04 | 1,763 | |
| DivPrunePruning Mode=T, Pruning Ratio=0.46, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=false, TFLOPs=3.012026.04 | 1,762 | |
| PyramidDropRetained Tokens=128, Pruning Ratio=↓ 77.8%2026.04 | 1,761 | |
| SparseVLMVTR Type=Text-aware, Tokens Retained=128, Reduction Ratio=22.2%2026.06 | 1,761 | |
| HoloVVTR Type=Text-agnostic, Tokens Retained=128, Reduction Ratio=22.2%2026.06 | 1,761 | |
| VisionZipPruning Mode=T, Pruning Ratio=0.62, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=true, TFLOPs=2.162026.04 | 1,756 | |
| SparseVLMBase Model=LLaVA-NeXT-13B, Token Retention Budget=160 Tokens2026.04 | 1,755 | |
| VisionZipVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,749 | |
| DeSAPRetained Tokens=64, Pruning Ratio=↓ 88.9%2026.04 | 1,748 | |
| DivPrune + RESTOREVTR Type=Text-agnostic, Tokens Retained=64, Reduction Ratio=11.1%2026.06 | 1,748 | |
| HiPrune++Backbone=LLaVA-1.5-13B, Token Num=1282025.08 | 1,745 | |
| FlowCutRetained Tokens=64, Pruning Ratio=↓ 88.9%2026.04 | 1,744 | |
| HiPrune++Backbone=LLaVA-1.5-13B, Token Num=642025.08 | 1,736 | |
| Baseline: all KVVLM Backbone=Qwen3-VL-8B-Instruct2026.05 | 1,735.6 | |
| CoMPPruning Mode=C, Pruning Ratio=0.62, Backbone=LLaVA-v1.5-7B, Supervised fine-tuning=true, TFLOPs=2.072026.04 | 1,735 | |
| HiPruneBackbone=LLaVA-1.5-13B, Token Num=1282025.08 | 1,730 | |
| ToMeVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,727 | |
| VisionZip + RESTOREVTR Type=Text-agnostic, Tokens Retained=64, Reduction Ratio=11.1%2026.06 | 1,726 | |
| CLASPBackbone=LLaVA-NeXT 7B, Tokens Retained=320, Pruning Ratio=88.9%2026.04 | 1,723 | |
| DivPruneVTR Type=Text-agnostic, Tokens Retained=192, Reduction Ratio=33.3%2026.06 | 1,723 | |
| VisPruner + RESTOREVTR Type=Text-agnostic, Tokens Retained=64, Reduction Ratio=11.1%2026.06 | 1,722 | |
| SparseVLMRetained Tokens=192, Pruning Ratio=↓ 66.7%2026.04 | 1,721 | |
| HoloVRetained Tokens=64, Pruning Ratio=↓ 88.9%2026.04 | 1,715 | |
| V2DropRetained Tokens=128, Pruning Ratio=↓ 77.8%2026.04 | 1,712 |