Probability-Entropy Calibration: An Elastic Indicator for Adaptive Fine-tuning

Token-level reweighting is a simple yet effective mechanism for controlling supervised finetuning, but common indicators are largely onedimensional: the ground-truth probability reflects downstream alignment, while token entropy reflects intrinsic uncertainty induced by the pretraining prior. Ignoring entropy can misidentify noisy or easily replaceable tokens as learningcritical, while ignoring probability fails to reflect target-specific alignment. RANKTUNER introduces a probability-entropy calibration signal, the Relative Rank Indicator, which compares the rank of the ground-truth token with its expected rank under the prediction distribution. The inverse indicator is used as a tokenwise Relative Scale to reweight the fine-tuning objective, focusing updates on truly under-learned tokens without over-penalizing intrinsically uncertain positions. Experiments on multiple backbones show consistent improvements on mathematical reasoning benchmarks, transfer gains on out-of-distribution reasoning, and pre code generation performance over probability-or entropyonly reweighting baselines. The implementation code is available at https://github.com/ LvAoAo/Ranktuner_VERL .