Reward Performance on Twenty Questions High-data
5.83Test-Time RewardILQL
Evaluation Results
| Method | Links | |
|---|---|---|
| ILQLtraining_regime=BOOST2026.05 | 5.83 | |
| ILQLtraining_regime=Bilevel Only2026.05 | 5.38 | |
| MCtraining_regime=Bilevel Only2026.05 | 3.83 | |
| MCtraining_regime=BOOST2026.05 | 3.48 | |
| ILQLtraining_regime=Vanilla2026.05 | 2.6 | |
| MCtraining_regime=Vanilla2026.05 | 2.22 | |
| MCtraining_regime=Vanilla+Syn2026.05 | 2.19 | |
| ILQLtraining_regime=Vanilla+Syn2026.05 | 2.04 |