Model Alignment on HH-RLHF D3 (test)
32.77Harmlessness BLEU ScoreDEFT-PRO
Evaluation Results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| DEFT-PROtrained_on=D32026.04 | 32.77 | 3.79 | 73.79 | 34.66 | 3.65 | 71.24 | 34.15 | 3.69 | 71.93 | |
| DEFT-DPOtrained_on=D32026.04 | 32.03 | 3.95 | 71.45 | 36.77 | 4.16 | 73.12 | 35.49 | 4.1 | 72.67 | |
| SFTtrained_on=D32026.04 | 31.76 | 3.86 | 72.48 | 34.91 | 3.84 | 68.54 | 34.06 | 3.85 | 69.6 | |
| PROtrained_on=D32026.04 | 29.4 | 3.56 | 72.95 | 33.5 | 3.64 | 68.49 | 33.5 | 3.62 | 69.69 | |
| DPOtrained_on=D32026.04 | 29.03 | 3.88 | 74.23 | 34.79 | 4.04 | 69.27 | 33.23 | 4 | 70.61 |