ResearchBenchmarksFine-tuning ranking evaluation on English fine-tuning corpus 30B tokensFollow0.847Observed LossLinear-H0.84660.84930.8520.8547Sep 7, 2024Evaluation ResultsMethodMethodLinksObserved LossPredicted ScoreActual RankPredicted RankLinear-HSchedule=Linear-H, ηma...Schedule=Linear-H, ηmax=1.5×10−3, a1=128, a3=1282024.090.8472.00411Linear-LSchedule=Linear-L, ηma...Schedule=Linear-L, ηmax=10−3, a1=512, a3=5122024.090.8492.02822Linear-SSchedule=Linear-S, ηma...Schedule=Linear-S, ηmax=5×10−4, a1=128, a3=1282024.090.8572.11633