List your company

Model Alignment on HH-RLHF D3 (test)

32.77Harmlessness BLEU ScoreDEFT-PRO
28.880429.890230.931.9098Apr 2, 2026

Evaluation results

MethodLinks
DEFT-PROtrained_on=D32026.0432.773.7973.7934.663.6571.2434.153.6971.93
DEFT-DPOtrained_on=D32026.0432.033.9571.4536.774.1673.1235.494.172.67
SFTtrained_on=D32026.0431.763.8672.4834.913.8468.5434.063.8569.6
PROtrained_on=D32026.0429.43.5672.9533.53.6468.4933.53.6269.69
DPOtrained_on=D32026.0429.033.8874.2334.794.0469.2733.23470.61