Multimodal Understanding on Aggregate Image and Video Benchmarks
100RelAccQ2.5VL-7B
Evaluation results
| Method | Links | |
|---|---|---|
| Q2.5VL-7BTokens Retained Percentage=100%2025.03 | 100 | |
| TwigVLM++Tokens Retained Percentage=33.3%2025.03 | 99.7 | |
| TwigVLM++Tokens Retained Percentage=22.2%2025.03 | 98.7 | |
| TwigVLMTokens Retained Percentage=33.3%2025.03 | 97.5 | |
| VisionZipTokens Retained Percentage=33.3%2025.03 | 97.2 | |
| FastVTokens Retained Percentage=33.3%2025.03 | 95.2 | |
| TwigVLMTokens Retained Percentage=22.2%2025.03 | 95 | |
| VisionZipTokens Retained Percentage=22.2%2025.03 | 94.7 | |
| TwigVLM++Tokens Retained Percentage=11.1%2025.03 | 94.4 | |
| FastVTokens Retained Percentage=22.2%2025.03 | 91.1 | |
| VisionZipTokens Retained Percentage=11.1%2025.03 | 88.4 | |
| TwigVLMTokens Retained Percentage=11.1%2025.03 | 86 | |
| FastVTokens Retained Percentage=11.1%2025.03 | 76.7 |