Reward Modeling on SafeRLHF Reversed
88.4AccuracyICRM
Evaluation Results
| Method | Links | |
|---|---|---|
| ICRMN (Number of in-context demonstrations)=322026.02 | 88.4 | |
| ICRMN (Number of in-context demonstrations)=162026.02 | 88.3 | |
| ICRMN (Number of in-context demonstrations)=82026.02 | 86.9 | |
| ICRMN (Number of in-context demonstrations)=42026.02 | 85.8 | |
| ICRMN (Number of in-context demonstrations)=22026.02 | 79.6 | |
| ICRMN (Number of in-context demonstrations)=12026.02 | 54.8 | |
| BTRM2026.02 | 12.8 | |
| ArmoRM2026.02 | 11.8 | |
| GRM2026.02 | 10.2 |