LLM Personalization on Anthropic HHH (test)
56.6Personalization WinPTDPO
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| PTDPOPersona Source=Pgold, Baseline Method=DPO2025.01 | 56.6 | 26 | 17.4 | 33.6 | 27.8 | 38.6 | 23 | |
| PTDPOPersona Source=Pretr, Baseline Method=DPO2025.01 | 55.8 | 25 | 19.2 | 25.4 | 25.2 | 49.4 | 8.4 | |
| PTSFTPersona Source=Pgold, Baseline Method=SFT2025.01 | 53.4 | 27.9 | 18.7 | 37.5 | 30.3 | 32.2 | 27.8 | |
| PTFSPersona Source=Pgold, Baseline Method=FS2025.01 | 49 | 18 | 33.1 | 43.7 | 17.3 | 39 | 12.5 | |
| PTSFTPersona Source=Pretr, Baseline Method=SFT2025.01 | 47.6 | 30.6 | 21.9 | 28.3 | 30.6 | 41.1 | 9.3 | |
| PTFSPersona Source=Pretr, Baseline Method=FS2025.01 | 46.6 | 18.3 | 35.1 | 38.4 | 15.6 | 46 | 2.5 |