Multimodal Understanding on LLaVA-NEXT-7B Evaluation Suite (test)
64.2GQA AccuracyVanilla
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| VanillaLoc.=-, Retained Tokens=2880, Pruning Rate=0%, Base Model=LLaVA-NEXT-7B2026.03 | 64.2 | 67.4 | 60.6 | 1,851 | 86.5 | 70.1 | 80.8 | 64.9 | 100 | |
| AutoSelectLoc.=Pre, Retained Tokens=320, Pruning Rate=88.9%, Base Model=LLaVA-NEXT-7B2026.03 | 62.3 | 64.7 | 57.4 | 1,723 | 85.9 | 72.7 | 78.6 | 56.7 | 96.1 | |
| DARTLoc.=L2, Retained Tokens=320, Pruning Rate=88.9%, Base Model=LLaVA-NEXT-7B2026.03 | 61.7 | 65.3 | 58.2 | 1,710 | 84.1 | 68.4 | 79.1 | 58.7 | 95.6 | |
| HoloVLoc.=Pre, Retained Tokens=320, Pruning Rate=88.9%, Base Model=LLaVA-NEXT-7B2026.03 | 61.7 | 65.3 | 57.5 | 1,738 | 83.9 | 68.9 | 79.5 | 58.7 | 95.7 | |
| HiREDLoc.=Pre, Retained Tokens=320, Pruning Rate=88.9%, Base Model=LLaVA-NEXT-7B2026.03 | 59.3 | 64.2 | 55.9 | 1,690 | 83.3 | 66.7 | 75.7 | 58.8 | 93.4 | |
| PDropLoc.=L8–L24, Retained Tokens=320, Pruning Rate=88.9%, Base Model=LLaVA-NEXT-7B2026.03 | 56.4 | 63.4 | 56.2 | 1,663 | 77.6 | 67.5 | 73.5 | 54.4 | 90.7 | |
| FastVLoc.=L2–L32, Retained Tokens=320, Pruning Rate=88.9%, Base Model=LLaVA-NEXT-7B2026.03 | 55.9 | 61.6 | 51.9 | 1,661 | 71.7 | 62.8 | 71.9 | 55.7 | 87.6 |