Multimodal Evaluation on MLLM Suite Subset (GQA, MMB, MME, POPE) (various)
60.84GQAQwen2.5-VL-7B
Evaluation Results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| Qwen2.5-VL-7BBackbone=Qwen2.5-VL-7B, Token Retention Rate=100%2025.09 | 60.84 | 84.1 | 2,310 | 86.3 | 100 | |
| DivPruneBackbone=Qwen2.5-VL-7B, Token Retention Rate=20%2025.09 | 60.05 | 79.55 | 2,173 | 83.42 | 96 | |
| ZOO-PruneBackbone=Qwen2.5-VL-7B, Token Retention Rate=20%2025.09 | 58.81 | 80.6 | 2,201 | 84.17 | 96.2 | |
| VisionZipBackbone=Qwen2.5-VL-7B, Token Retention Rate=20%2025.09 | 57.27 | 79.72 | 2,221 | 83.89 | 95.6 | |
| DivPruneBackbone=Qwen2.5-VL-7B, Token Retention Rate=10%2025.09 | 55.49 | 76.03 | 2,054 | 79.05 | 90.5 | |
| ZOO-PruneBackbone=Qwen2.5-VL-7B, Token Retention Rate=10%2025.09 | 55.45 | 76.28 | 2,018 | 80.99 | 90.8 | |
| VisionZipBackbone=Qwen2.5-VL-7B, Token Retention Rate=10%2025.09 | 54.09 | 76.03 | 1,937 | 78.97 | 88.7 |