This work empirically evaluates the primary performance-profile estimator on MATH-500, ScienceQA, MMLU, WinoGrande, HellaSwag, TruthfulQA, GSM8K, and ARC and proves calibration-free identification, unbiasedness for grouped profiles, local oracle optimality within centered linear augmentations, and first-order adaptivit...
Zhi Zhang, Lingfeng Lyu, Yue Kang et al.· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.