Visual Question Answering on GQA (GQA Score and Avg. Normalized Average Retention)
64.3GQA ScoreVanilla
Evaluation results
| Method | Links | ||
|---|---|---|---|
| VanillaModel=LLaVA-NeXT, TFLOPs=100%2026.06 | 64.3 | 100 | |
| VanillaBase Model=LLaVA-NEXT-7B, Token Budget=2880, Pruning Rate=0%, Loc. (Pruning Location)=-2026.05 | 64.2 | 100 | |
| VSkip+ (N=20)Model=LLaVA-NeXT, TFLOPs=74%2026.06 | 63.31 | 98.2 | |
| Ours (N=20)Model=LLaVA-NeXT, TFLOPs=67%2026.06 | 63.01 | 98.8 | |
| APETModel=LLaVA-NeXT, TFLOPs=70%2026.06 | 62.97 | 99.4 | |
| ShortV (N=20)Model=LLaVA-NeXT, TFLOPs=51%2026.06 | 62.83 | 96 | |
| V2DropModel=LLaVA-NeXT, TFLOPs=70%2026.06 | 62.58 | 98.4 | |
| DiffPruneBase Model=LLaVA-NEXT-7B, Token Budget=320, Pruning Rate=88.9%, Loc. (Pruning Location)=Pre2026.05 | 62.3 | 96.1 | |
| DARTBase Model=LLaVA-NEXT-7B, Token Budget=320, Pruning Rate=88.9%, Loc. (Pruning Location)=L22026.05 | 61.7 | 95.6 | |
| HoloVBase Model=LLaVA-NEXT-7B, Token Budget=320, Pruning Rate=88.9%, Loc. (Pruning Location)=Pre2026.05 | 61.7 | 95.7 | |
| HiREDBase Model=LLaVA-NEXT-7B, Token Budget=320, Pruning Rate=88.9%, Loc. (Pruning Location)=Pre2026.05 | 59.3 | 93.4 | |
| PDropBase Model=LLaVA-NEXT-7B, Token Budget=320, Pruning Rate=88.9%, Loc. (Pruning Location)=L8–L242026.05 | 56.4 | 90.7 | |
| FastVBase Model=LLaVA-NEXT-7B, Token Budget=320, Pruning Rate=88.9%, Loc. (Pruning Location)=L2–L322026.05 | 55.9 | 87.6 | |
| VTW (K=16)Model=LLaVA-NeXT, TFLOPs=51%2026.06 | 55.41 | 86.5 |