mSFT: Addressing Dataset Mixtures Overfiting Heterogeneously in Multi-task SFTMar 25, 2026·[P1] Woosung Koh*,Jeyoung Jeon*,Youngjin Song,Yujin Cheon,Soowon Oh,Jaehyeong Choi,Se-Young Yun· 0 min read PDF CodeTypeConference paperPublicationPre-printLast updated on May 2, 2026Post-Training Multi-Task Generalization ← Large Language Models Can Control Their Own Attention Span May 22, 2026A Taxonomy of Heterogeneous Statistical Interdependencies for Graph-based Financial Time Series Prediction Feb 1, 2026 →