Personalized Reward Modeling on PRISM Seen
65.3User-level AccuracyMRM
Evaluation results
| Method | Links | |
|---|---|---|
| MRMBase Reward Model=Skywork-Reward V22026.01 | 65.3 | |
| MRMBase Reward Model=Skywork-Reward V12026.01 | 64.8 | |
| VPL2026.01 | 64.6 | |
| BT2026.01 | 64.3 | |
| SynthesizeMeAdaptation Protocol=FT2026.01 | 64.3 | |
| GPO2026.01 | 64.2 | |
| SynthesizeMeAdaptation Protocol=ICL2026.01 | 63.9 | |
| LoRe2026.01 | 63 | |
| PAL2026.01 | 61.7 | |
| Skywork-Reward V2Base Reward Model=V22026.01 | 60.5 | |
| Skywork-Reward V1Base Reward Model=V12026.01 | 59.5 |