DIRECT: Direct Decoding for Efficient and Aligned Sequence Labeling with Large Language Models

Yilei Wang; Jiaxin Gan; Kexuan Zhang; Ling Li; Wentao Zhang; Peichao Lai

DIRECT: Direct Decoding for Efficient and Aligned Sequence Labeling with Large Language Models

Computation and Language 2026-07-29 v1

Authors: Yilei Wang , Jiaxin Gan , Kexuan Zhang , Ling Li , Wentao Zhang , Peichao Lai

Abstract

Sequence labeling is a fine-grained information extraction task, yet existing large language model-based approaches suffer from insufficient domain alignment and low inference efficiency. To address these issues, we propose DIRECT, a framework that addresses these issues through training-time optimization and inference-time rectification. Specifically, DIRECT 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. To further improve efficiency, a template-filling mechanism requires the model to generate only label tokens while reusing prefixed content through the KV Cache, thus reducing redundant computation. Experimental results on eight datasets demonstrate that DIRECT achieves significant improvements in both performance and efficiency compared to existing methods.

Cite

@article{arxiv.2607.26891,
  title  = {DIRECT: Direct Decoding for Efficient and Aligned Sequence Labeling with Large Language Models},
  author = {Yilei Wang and Jiaxin Gan and Kexuan Zhang and Ling Li and Wentao Zhang and Peichao Lai},
  journal= {arXiv preprint arXiv:2607.26891},
  year   = {2026}
}