ResearchBenchmarksReasoning and Generative Tasks on MMLU, CMMLU, GSM8k, XSum, and StrategyQA (test)Follow30.77MMLU AccuracyGradMAP25.20626.650528.09529.5395Feb 16, 2026Evaluation ResultsMethodMethodLinksMMLU AccuracyCMMLU AccuracyGSM8k AccuracyXSum ScoreStrategyQA AccuracyAverage ScoreGradMAPBackbone=Baichuan2-7B,...Backbone=Baichuan2-7B, Variant=Entry 12026.0230.7748.8911.451.6740.7426.7GradMAPBackbone=Baichuan2-7B,...Backbone=Baichuan2-7B, Variant=Entry 22026.0230.1751.8411.512.1240.7927.2ShortGPTBackbone=Baichuan2-7BBackbone=Baichuan2-7B2026.0228.5940.627.891.6720.1319.78SLEBBackbone=Baichuan2-7BBackbone=Baichuan2-7B2026.0225.4222.360.890.1518.4713.46