Multimodal Understanding across MMB, MMBC, MMStar, MMMU, and AI2D
71.2Average AccuracyQwen2.5-VL-7B
Evaluation Results
| Method | Links | |
|---|---|---|
| Qwen2.5-VL-7BToken Retention Ratio=100%, Backbone=Qwen2.5-VL-7B2026.02 | 71.2 | |
| EntropyPruneToken Retention Ratio=25%, Backbone=Qwen2.5-VL-7B2026.02 | 66.9 | |
| FastVToken Retention Ratio=25%, Backbone=Qwen2.5-VL-7B2026.02 | 63.8 | |
| CDPrunerToken Retention Ratio=25%, Backbone=Qwen2.5-VL-7B2026.02 | 62.9 | |
| EntropyPruneToken Retention Ratio=12.5%, Backbone=Qwen2.5-VL-7B2026.02 | 62.4 | |
| CDPrunerToken Retention Ratio=12.5%, Backbone=Qwen2.5-VL-7B2026.02 | 60.1 | |
| FastVToken Retention Ratio=12.5%, Backbone=Qwen2.5-VL-7B2026.02 | 49.8 |