Visual Question Answering on ScienceQA SQA-IMG
74.4SQAIMG AccuracyDART
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| DARTBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 74.4 | 97.1 | |
| SparseVLMBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 74.1 | 97.9 | |
| DARTBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 73.9 | 92.5 | |
| SPAREBackbone=LLaVA-1.5-13B, Visual Token Retention=32 tokens2026.06 | 73.2 | 94.7 | |
| LLaVA-1.5-13BBackbone=LLaVA-1.5-13B, Visual Token Retention=576 tokens2026.06 | 72.8 | 100 | |
| SPAREBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 72.8 | 98 | |
| DARTBackbone=LLaVA-1.5-13B, Visual Token Retention=32 tokens2026.06 | 72.8 | 86 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 72.6 | 97.9 | |
| SparseVLMBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 72.5 | 89.7 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 72.5 | 96.3 | |
| SPAREBackbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 72.3 | 97.2 | |
| CDPrunerBackbone=LLaVA-1.5-13B, Visual Token Retention=32 tokens2026.06 | 71.9 | 92.6 | |
| PruMerge+Backbone=LLaVA-1.5-13B, Visual Token Retention=128 tokens2026.06 | 71.8 | 93.2 | |
| PruMerge+Backbone=LLaVA-1.5-13B, Visual Token Retention=64 tokens2026.06 | 71.7 | 90 | |
| PyramidDropModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.05 | 69.5 | 90.1 | |
| DARTModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.05 | 68.3 | 96.4 | |
| SPprunerModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.05 | 68 | 97.7 | |
| DivPruneModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.05 | 67.7 | 96.1 | |
| VanillaModel=LLaVA-Next-7B, Token Budget=2880, Reduction Ratio=0%2026.05 | 67.6 | 100 | |
| VisionZipModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.05 | 67.6 | 96.2 | |
| FastVModel=LLaVA-Next-7B, Token Budget=320, Reduction Ratio=88.9%2026.05 | 62.8 | 86.8 |