Multimodal Evaluation on MME (MME, Rel. Acc)
1,534.8MME ScoreDART
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DARTBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 1,534.8 | 97.1 | |
| LLaVA-1.5-13BBackbone=LLaVA-1.5-13B, Visual Token Retention=576 tokens2026.06 | 1,531.2 | 100 | |
| SparseVLMBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 1,494.3 | 97.9 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 1,468.5 | 97.9 | |
| SPAREBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 1,467.4 | 97.2 | |
| SPAREBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 1,449.8 | 98 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 1,436.7 | 96.3 | |
| DARTBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 1,433 | 92.5 | |
| SPAREBackbone=LLaVA-1.5-13B, Visual Token Retention=32 tokens2026.06 | 1,420.9 | 94.7 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Visual Token Retention=32 tokens2026.06 | 1,417.3 | 92.6 | |
| PruMerge+Backbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 1,403.2 | 93.2 | |
| SparseVLMBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 1,353.5 | 89.7 | |
| PruMerge+Backbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 1,332.5 | 90 | |
| DARTBackbone=LLaVA-1.5-13B, Visual Token Retention=32 tokens2026.06 | 1,319.7 | 86 |