Reward Performance on Twenty Questions Low-data
3.46Test-Time RewardMC
Evaluation Results
| Method | Links | |
|---|---|---|
| MCtraining_regime=BOOST2026.05 | 3.46 | |
| MCtraining_regime=Bilevel Only2026.05 | 2.9 | |
| MCtraining_regime=Vanilla2026.05 | 1.01 | |
| MCtraining_regime=Vanilla+Syn2026.05 | 1.01 | |
| ILQLtraining_regime=BOOST2026.05 | 0.53 | |
| ILQLtraining_regime=Bilevel Only2026.05 | 0.51 | |
| ILQLtraining_regime=Vanilla+Syn2026.05 | 0.45 | |
| ILQLtraining_regime=Vanilla2026.05 | 0.38 |