Vision-Language Evaluation on MME
2,498MME ScoreVanilla
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VanillaBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=100%2026.05 | 2,498 | — | — | |
| VisionZipBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=70%2026.05 | 2,475 | — | — | |
| VisMMOEBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=70%2026.05 | 2,470 | — | — | |
| MixBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=70%2026.05 | 2,375 | — | — | |
| VisionZipBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=50%2026.05 | 2,309 | — | — | |
| VisMMOEBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=50%2026.05 | 2,299 | — | — | |
| VanillaBackbone=DeepSeek-VL2, Pruning Ratio=100%2026.05 | 2,243 | — | — | |
| MixBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=50%2026.05 | 2,212 | — | — | |
| VisionZipBackbone=DeepSeek-VL2, Pruning Ratio=70%2026.05 | 2,195 | — | — | |
| RandomBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=70%2026.05 | 2,181 | — | — | |
| VisMMOEBackbone=DeepSeek-VL2, Pruning Ratio=70%2026.05 | 2,178 | — | — | |
| MixBackbone=DeepSeek-VL2, Pruning Ratio=70%2026.05 | 2,153 | — | — | |
| VisionZipBackbone=DeepSeek-VL2, Pruning Ratio=50%2026.05 | 2,142 | — | — | |
| RandomBackbone=DeepSeek-VL2, Pruning Ratio=70%2026.05 | 2,115 | — | — | |
| VisMMOEBackbone=DeepSeek-VL2, Pruning Ratio=50%2026.05 | 2,103 | — | — | |
| MixBackbone=DeepSeek-VL2, Pruning Ratio=50%2026.05 | 2,090 | — | — | |
| RandomBackbone=Qwen3-VL-30B-A3B, Pruning Ratio=50%2026.05 | 2,053 | — | — | |
| RandomBackbone=DeepSeek-VL2, Pruning Ratio=50%2026.05 | 2,014 | — | — | |
| VanillaBase Model=LLaVA-1.5-7B2025.05 | 1,862 | — | — | |
| MoBBase Model=LLaVA-1.5-7B, Pruning budget K=192, η-prior=true2025.05 | 1,860 | — | — | |
| VanillaBase Model=LLaVA-Next-7B2025.05 | 1,851 | — | — | |
| Upper BoundBackbone=LLaVA-v1.5-13B, Retained Tokens=Full, Pruning Ratio=0%2026.04 | 1,818 | — | — | |
| MoBBase Model=LLaVA-1.5-7B, Pruning budget K=64, η-prior=true2025.05 | 1,806 | — | — | |
| CLASPBackbone=LLaVA-v1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.04 | 1,789 | — | — | |
| CLASPBackbone=LLaVA-v1.5-13B, Retained Tokens=192, Pruning Ratio=66.7%2026.04 | 1,785 | — | — | |
| SparseVLMBackbone=LLaVA-v1.5-13B, Retained Tokens=128, Pruning Ratio=77.8%2026.04 | 1,774 | — | — | |
| SparseVLMBackbone=LLaVA-v1.5-13B, Retained Tokens=192, Pruning Ratio=66.7%2026.04 | 1,768 | — | — | |
| MoBBase Model=LLaVA-Next-7B, Pruning budget K=320, η-prior=true2025.05 | 1,760 | — | — | |
| CLASPBackbone=LLaVA-v1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.04 | 1,675 | — | — | |
| SparseVLMBackbone=LLaVA-v1.5-13B, Retained Tokens=64, Pruning Ratio=88.9%2026.04 | 1,641 | — | — | |
| CLASPBackbone=LLaVA-v1.5-13B, Retained Tokens=32, Pruning Ratio=94.4%2026.04 | 1,592 | — | — | |
| BaselineRetain Tokens=Full2026.02 | 1,463 | — | — | |
| SparseVLMBackbone=LLaVA-v1.5-13B, Retained Tokens=32, Pruning Ratio=94.4%2026.04 | 1,451 | — | — | |
| IDPrunerRetain Tokens=1282026.02 | 1,428 | — | — | |
| IDPrunerRetain Tokens=642026.02 | 1,329 | — | — | |
| IDPrunerRetain Tokens=322026.02 | 1,254 | — | — | |
| BLIP-22024.05 | — | 1,293.8 | 290 | |
| IDEFICS2024.05 | — | 1,177.3 | — | |
| InstructBLIP2024.05 | — | 1,212.8 | 291.8 | |
| LLaVA-1.5-7BSIMA=false2024.05 | — | 1,506.4 | 355.7 | |
| LLaVA-1.5-7BSIMA=true2024.05 | — | 1,507.7 | 379.3 | |
| mPLUG-Owl22024.05 | — | 1,450.2 | 313.2 | |
| Qwen-VL-Chat2024.05 | — | 1,487.6 | 360.7 | |
| VILA-7BSIMA=false2024.05 | — | 1,533 | 316.4 | |
| VILA-7BSIMA=true2024.05 | — | 1,559.6 | 326.8 |