Text-centric Visual Question Answering on VQAT (Score)
73.8ScoreNEO-2B
Evaluation Results
| Method | Links | |
|---|---|---|
| NEO-2BBackbone=NEO-2B, Token Retention Ratio=100%, Resolution=Dynamic Resolution2026.06 | 73.8 | |
| ERABackbone=NEO-2B, Token Retention Ratio=50%, Resolution=Dynamic Resolution2026.06 | 72.4 | |
| CDPrunerBackbone=NEO-2B, Token Retention Ratio=50%, Resolution=Dynamic Resolution2026.06 | 71.2 | |
| ERABackbone=NEO-2B, Token Retention Ratio=20%, Resolution=Dynamic Resolution2026.06 | 68 | |
| DivPruneBackbone=NEO-2B, Token Retention Ratio=50%, Resolution=Dynamic Resolution2026.06 | 67.2 | |
| CDPrunerBackbone=NEO-2B, Token Retention Ratio=20%, Resolution=Dynamic Resolution2026.06 | 62.2 | |
| ERABackbone=NEO-2B, Token Retention Ratio=10%, Resolution=Dynamic Resolution2026.06 | 60.5 | |
| FastVBackbone=NEO-2B, Token Retention Ratio=50%, Resolution=Dynamic Resolution2026.06 | 53.7 | |
| CDPrunerBackbone=NEO-2B, Token Retention Ratio=10%, Resolution=Dynamic Resolution2026.06 | 52.9 | |
| FastVBackbone=NEO-2B, Token Retention Ratio=20%, Resolution=Dynamic Resolution2026.06 | 44.9 | |
| DivPruneBackbone=NEO-2B, Token Retention Ratio=20%, Resolution=Dynamic Resolution2026.06 | 41.1 | |
| FastVBackbone=NEO-2B, Token Retention Ratio=10%, Resolution=Dynamic Resolution2026.06 | 30.8 | |
| DivPruneBackbone=NEO-2B, Token Retention Ratio=10%, Resolution=Dynamic Resolution2026.06 | 23.3 |