Vision-Language Reasoning on BEAF (test)
88.4Simple AccuracyQ4 system redistr (prop)
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Q4 system redistr (prop)Intervention=Q4 system redistribution, strategy=proportional2026.01 | 88.4 | 84.74 | 32.62 | |
| Image×2.0Intervention=Image×2.02026.01 | 85.73 | 81.01 | 41.55 | |
| Q4 text redistr (prop)Intervention=Q4 text redistribution, strategy=proportional2026.01 | 85.55 | 80.75 | 41.68 | |
| Q4 system ablIntervention=Q4 system ablation2026.01 | 84.44 | 79.45 | 43.58 | |
| No intervention baselineIntervention=None2026.01 | 83.74 | 78.62 | 44.69 | |
| AD-HHIntervention=AD-HH2026.01 | 83.67 | 78.51 | 44.74 | |
| PAIIntervention=PAI2026.01 | 83.01 | 77.72 | 45.73 |