Calibration

MMLU-Redux → AA-HLE

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap

AA-HLE is estimated from MMLU-Redux with a offset logistic curve fitted on 7 models measured on both: y = 0.0888 + (0.2720 − 0.0888) / (1 + exp(−200.00·(x − 0.9251))), R² = 0.35, cross-validated error 11.5 pp. It is used for 0 estimates.