Preference Evaluation on PKU-SafeRLHF
57Win RateDLMA-13B
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DLMA-13BComparison Baseline=Llama2-13B2024.02 | 57 | 8 | 35 | |
| DLMA-7BComparison Baseline=RLAIF-7B2024.02 | 56 | 8 | 36 | |
| DLMA-7BComparison Baseline=Llama2-7B2024.02 | 55 | 8 | 37 | |
| DLMA-13BComparison Baseline=RLAIF-13B2024.02 | 55 | 11 | 34 | |
| DLMA-13BComparison Baseline=CD-13B2024.02 | 49 | 16 | 45 | |
| DLMA-7BComparison Baseline=RLCD-7B2024.02 | 43 | 25 | 32 | |
| DLMA-13BComparison Baseline=RLCD-13B2024.02 | 43 | 24 | 33 | |
| DLMA-7BComparison Baseline=CD-7B2024.02 | 42 | 15 | 43 |