Vision-Language Multi-task Evaluation on Qwen2.5-VL Suite (test)
82.8MMB ScoreVanilla
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| VanillaPruning Rate=100%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 82.8 | 2,304 | 86.1 | 84.7 | 84.8 | 100 | |
| DiffPrunePruning Rate=66.7%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 81.7 | 2,279 | 84.9 | 86.4 | 79 | 98.3 | |
| DiffPrunePruning Rate=77.8%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 81 | 2,218 | 82.8 | 83.7 | 76.7 | 95.9 | |
| HoloVPruning Rate=66.7%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 78.3 | 2,093 | 85 | 79.8 | 78.9 | 94.3 | |
| DiffPrunePruning Rate=88.9%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 76.7 | 2,113 | 79.8 | 82.9 | 76.7 | 93.1 | |
| HoloVPruning Rate=77.8%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 76.5 | 2,043 | 82.3 | 79.8 | 70.3 | 90.8 | |
| FastVPruning Rate=66.7%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 75.7 | 2,072 | 82.2 | 78.5 | 77.9 | 92.3 | |
| FastVPruning Rate=77.8%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 74.9 | 2,036 | 80.7 | 78 | 69 | 89.2 | |
| HoloVPruning Rate=88.9%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 72.4 | 2,006 | 80.7 | 79.5 | 61.8 | 87 | |
| FastVPruning Rate=88.9%, Backbone=Qwen2.5-VL-7B, Input Resolution=Native-resolution2026.05 | 69.2 | 1,940 | 78.6 | 77.4 | 60.3 | 84.3 |