Jul 2026
DIRECT: Direct Decoding for Efficient and Aligned Sequence Labeling with Large Language Models
DI performs Direct Preference Optimization (DPO) after supervised fine-tuning to strengthen task alignment with human preferences, and introduces a controlled decoding process that enforces fixed output formats and restricts predictions to candidate sets.
Yilei Wang, Jiaxin Gan, Kexuan Zhang et al.
· arXiv.org · 0 citations