FamiFT: Familiarity-Adaptive Supervised Fine-Tuning

Yue Wu, Yangrong Hu, Xueyu Zhou, Jian Huang.

Under review at ICLR 2027.

Overview

Familiarity-adaptive token weighting based on pretrained-model familiarity improves mean mathematical reasoning accuracy from 21.32% to 37.20% over standard SFT on Qwen2.5-Math-7B.

All publications & manuscripts →