Multimodal Question Answering on LLaVA-NeXT-7B Evaluation Suite
5.6FLOPs (TFLOPs)VisionZip
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| VisionZipToken (#)=1602026.06 | 5.6 | 232 | 93 | 33 | 0.08 | 15.3 | 88.9 | |
| CDPrunerToken (#)=1602026.06 | 5.6 | 198 | 85 | 28 | 0.08 | 14.3 | 95.5 | |
| ERAToken (#)=1602026.06 | 5.6 | 200 | 91 | 27 | 0.08 | 14.6 | 95.8 | |
| DARTToken (#)=1602026.06 | 8.6 | 192 | 93 | 27 | 0.08 | 13.8 | 90.5 | |
| PDropToken (#)=1602026.06 | 17.4 | 249 | 129 | 29 | 0.08 | 16.1 | 79.2 | |
| LLaVA-NeXT-7BToken (#)=28802026.06 | 66.1 | 544 | 415 | 25 | 1.44 | 18.6 | 100 |