Personalized Reward Modeling on LaMP 4
17.9ROUGE@N=30Probabilistic User RM
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Probabilistic User RMCost/query=0.05 s2026.05 | 17.9 | — | 15.5 | |
| LLM tournament (Qwen3-4B)Cost/query=1.63 s2026.05 | 15.7 | 49.9 | 1.3 | |
| Random2026.05 | 15.5 | 50 | — | |
| LLM pointwise (GPT-class)Cost/query=30 calls2026.05 | 14.4 | — | 7.1 |