Multi-turn Visual Question Answering on MT-GQA
96.9Time To First Token (TTFT)Sample
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| SampleModel=LLaVA-1.5-7b, Reduction Rate=90%2026.03 | 96.9 | 482 | 26.2 | 13.3 | |
| MetaCompressModel=LLaVA-1.5-7b, Reduction Rate=90%2026.03 | 97.8 | 480 | 26.1 | 13.3 | |
| RandomModel=LLaVA-1.5-7b, Reduction Rate=90%2026.03 | 98.2 | 487 | 26.2 | 13.3 | |
| FastVModel=LLaVA-1.5-7b, Reduction Rate=90%2026.03 | 102 | 528 | 26.3 | 13.5 | |
| PruMergeModel=LLaVA-1.5-7b, Reduction Rate=90%2026.03 | 107 | 509 | 26.2 | 13.3 | |
| RandomModel=LLaVA-NeXT-7b, Reduction Rate=90%2026.03 | 174 | 481 | 14.8 | 12.7 | |
| MetaCompressModel=LLaVA-NeXT-7b, Reduction Rate=90%2026.03 | 174 | 501 | 14.9 | 12.7 | |
| SampleModel=LLaVA-NeXT-7b, Reduction Rate=90%2026.03 | 176 | 484 | 14.8 | 12.7 | |
| FastVModel=LLaVA-NeXT-7b, Reduction Rate=90%2026.03 | 219 | 529 | 19.2 | 12.9 | |
| BaseModel=LLaVA-1.5-7b, Reduction Rate=0%2026.03 | 232 | 676 | 26.9 | 71.4 | |
| BaseModel=LLaVA-NeXT-7b, Reduction Rate=0%2026.03 | 484 | 830 | 16.7 | 95.3 |