Multimodal Question Answering on ScienceQA (SQA) (test)
70.2SQA AccuracyVanilla (LLaVA-NeXT-7B)
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| Vanilla (LLaVA-NeXT-7B)#Tokens=28802026.01 | 70.2 | — | — | 9.6 | — | 1,512.1 | — | |
| VisionTrim#Tokens=3202026.01 | 69.6 | — | 61.4 | 0.8 | 91.7 | 101.8 | 93.3 | |
| VisionZip#Tokens=3202026.01 | 67.3 | — | 32.7 | 1.6 | 83.3 | 180.4 | 88.1 | |
| SparseVLM#Tokens=3202026.01 | 67 | — | 30.6 | 1.5 | 84.4 | 168 | 88.9 |