ResearchBenchmarksQuestion Answering on ARC-C Small Shift (test)Follow69.5AccuracyBayesian-LoRA67.21267.80668.468.994Jan 28, 2026Evaluation ResultsMethodMethodLinksAccuracyExpected Calibration ErrorNegative Log LikelihoodBayesian-LoRASample size (S)=4, Tra...Sample size (S)=4, Training protocol=End-to-End2026.0169.58.10.82LA2026.0169.25.50.85LLLA2026.0168.121.30.88Temp2026.01689.410.9MAP2026.0167.922.21.3Ckpt Ens2026.0167.917.71.03Dropout2026.0167.721.41.24BLoBEnsemble size (N)=10Ensemble size (N)=102026.0167.79.550.83BBB2026.0167.319.91.06