Multimodal Understanding Inference Efficiency on InternVL2.5-8B
56.89Reduction RatioLUVC
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| LUVCreduction_strategy=Linear-time Universal Visual-token Compression2025.12 | 56.89 | 214 | 18.68 | |
| PACTreduction_strategy=PACT2025.12 | 42 | 155 | 19.68 | |
| VTWreduction_strategy=Visual Token Waltz2025.12 | 40.25 | 140 | 19.54 | |
| No reductionreduction_strategy=None2025.12 | 0 | 100 | 21.79 |