Identity Internalization on Identity Persona-axis
66.7Best ID Self-Name ConsistencyGuided OPSD
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Guided OPSDPersona source=user2026.05 | 66.7 | 68.8 | |
| RLSD-no-verifierGuidance=guided2026.05 | 62.5 | 64.6 | |
| EDGE-OPDAblation=w/o KL, Persona source=system2026.05 | 58.3 | 56.2 | |
| Guided OPSDPersona source=system2026.05 | 56.2 | 54.2 | |
| EDGE-OPDPersona source=user2026.05 | 56.2 | 58.3 | |
| EDGE-OPDAblation=w/o KL, Persona source=user2026.05 | 54.2 | 52.1 | |
| EDGE-OPDPersona source=system2026.05 | 52.1 | 50 | |
| Nemotron-3-Nano-4BDescription=Base model2026.05 | 0 | 0 | |
| OPSDGuidance=none2026.05 | 0 | 0 | |
| RLSD-no-verifierGuidance=unguided2026.05 | 0 | 0 |