Vision-Language Reasoning on HallusionBench (test)
53.31Simple AccuracyQ4 system redistr (prop)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Q4 system redistr (prop)Intervention=Q4 system redistribution, strategy=proportional2026.01 | 53.31 | 19.07 | 59.62 | |
| Q4 text redistr (prop)Intervention=Q4 text redistribution, strategy=proportional2026.01 | 46.58 | 9.77 | 84.23 | |
| Image×2.0Intervention=Image×2.02026.01 | 45.95 | 10 | 88.22 | |
| Q4 system ablIntervention=Q4 system ablation2026.01 | 45.74 | 10 | 89.27 | |
| PAIIntervention=PAI2026.01 | 45.64 | 9.53 | 90.22 | |
| No intervention baselineIntervention=None2026.01 | 45.64 | 9.77 | 89.8 | |
| AD-HHIntervention=AD-HH2026.01 | 45.22 | 9.3 | 90.43 |