Document Writing on Doc Writing
45.16BLEU ScoreVM+Norm-ITPO
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| VM+Norm-ITPOAdvantage Estimator=PPO, Credit Assignment=VM+Norm-ITPO2026.03 | 45.16 | 0.96 | 40.12 | |
| Norm-ITPOAdvantage Estimator=RLOO, Credit Assignment=Norm-ITPO2026.03 | 44.83 | 1.04 | 39.59 | |
| ITPOAdvantage Estimator=GRPO, Credit Assignment=ITPO2026.03 | 44.6 | 0.94 | 39.92 | |
| ITPOAdvantage Estimator=RLOO, Credit Assignment=ITPO2026.03 | 44.59 | 1.08 | 39.12 | |
| Norm-ITPOAdvantage Estimator=GRPO, Credit Assignment=Norm-ITPO2026.03 | 42.78 | 0.98 | 37.86 | |
| VM+ITPOAdvantage Estimator=PPO, Credit Assignment=VM+ITPO2026.03 | 42.52 | 1.02 | 37.44 | |
| LLM as JudgeAdvantage Estimator=GRPO, Credit Assignment=LLM as Judge2026.03 | 42.47 | 1.06 | 37.15 | |
| LLM as JudgeAdvantage Estimator=RLOO, Credit Assignment=LLM as Judge2026.03 | 42.34 | 1.04 | 37.2 | |
| VM+PRIMEAdvantage Estimator=PPO, Credit Assignment=VM+PRIME2026.03 | 41.87 | 2.42 | 29.77 | |
| PRIMEAdvantage Estimator=RLOO, Credit Assignment=PRIME2026.03 | 40.95 | 0.74 | 37.21 | |
| VM OnlyAdvantage Estimator=PPO, Credit Assignment=VM Only2026.03 | 40.57 | 0.84 | 36.37 | |
| PRIMEAdvantage Estimator=GRPO, Credit Assignment=PRIME2026.03 | 39.54 | 1.1 | 33.99 | |
| Trajectory-LevelAdvantage Estimator=GRPO, Credit Assignment=Trajectory-Level2026.03 | 39.33 | 0.36 | 37.28 | |
| Uniform DecomposeAdvantage Estimator=GRPO, Credit Assignment=Uniform Decompose2026.03 | 38.47 | 0.48 | 36 | |
| Uniform DecomposeAdvantage Estimator=RLOO, Credit Assignment=Uniform Decompose2026.03 | 37.81 | 0.86 | 33.45 | |
| Trajectory-LevelAdvantage Estimator=RLOO, Credit Assignment=Trajectory-Level2026.03 | 37.35 | 0.42 | 35.34 | |
| VanillaAdvantage Estimator=N/A, Credit Assignment=Vanilla2026.03 | 24.97 | 2.4 | 12.93 |