arXiv

ProFit: Leveraging High-Value Signals in SFT via Probability-Guided Token Selection

Tao Liu, Taiqiang Wu, Runming Yang

Jan 19, 2026·08:57·

00:00

08:57

Supervised Fine-Tuning (SFT)Single-Reference OverfittingToken Probability and Semantic ImportanceProFit MethodHigh-Value SignalsProFit method

Turn any paper into a podcast

ResearchPod turns research papers into podcasts you can actually follow.