Vision-Language Understanding on Vision-Language Benchmark Suite Aggregate
100Aggregate Performance (%)Vanilla
Evaluation Results
| Method | Links | |
|---|---|---|
| VanillaBackbone=LLaVA-1.5-7B, Retained Tokens=576, Compression Ratio=100%2026.04 | 100 | |
| EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%, Number of compressor layers (l)=22026.04 | 99.3 | |
| VisionZipBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 98.8 | |
| EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%, Number of compressor layers (l)=02026.04 | 98.7 | |
| DARTBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 98.3 | |
| EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%, Number of compressor layers (l)=22026.04 | 98 | |
| EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%, Number of compressor layers (l)=02026.04 | 97.8 | |
| DARTBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 97.5 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 97.4 | |
| VisionZipBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 97.4 | |
| PyramidDropBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 97 | |
| MustDropBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 97 | |
| FastVBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 96.9 | |
| HiREDBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 96.3 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 96.3 | |
| PyramidDropBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 95.9 | |
| MustDropBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 95.4 | |
| EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%, Number of compressor layers (l)=02026.04 | 94.9 | |
| HiREDBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 94.8 | |
| FastVBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 94.4 | |
| DARTBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%, Number of compressor layers (l)=02026.04 | 94.2 | |
| VisionZipBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 94.2 | |
| EvoCompBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%, Number of compressor layers (l)=22026.04 | 93.9 | |
| DARTBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 92.8 | |
| DARTBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%, Number of compressor layers (l)=02026.04 | 91.1 | |
| HiREDBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 90.8 | |
| PruMergeBackbone=LLaVA-1.5-7B, Retained Tokens=192, Compression Ratio=66.7%2026.04 | 90.7 | |
| MustDropBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 90 | |
| SparseVLMBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 89.4 | |
| PruMergeBackbone=LLaVA-1.5-7B, Retained Tokens=128, Compression Ratio=77.8%2026.04 | 88.6 | |
| FastVBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 87.6 | |
| PruMergeBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 86 | |
| PyramidDropBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%2026.04 | 84.9 | |
| DARTBackbone=LLaVA-1.5-7B, Retained Tokens=64, Compression Ratio=88.9%, Number of compressor layers (l)=02026.04 | 84.5 |