Hugging Peers

Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer Decoding

☆ Save
Zhoubian, Sining,Zhang, Xuanming,Chen, Yuxuan,Tang, Tianyi,Yang, An,Du, Sean,Zheng, Chujie,Huang, Fei
+3 more authors show fewer
Liu, Dayiheng,Huang, Gao,Zhou, Jingren
Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
published

Abstract

No abstract available.

Cite this paper

@article{sining2026,
  title = {Deeper is Not Always Better: Mitigating the Alignment Tax via Confident Layer Decoding},
  author = {Zhoubian, Sining and Zhang, Xuanming and Chen, Yuxuan and Tang, Tianyi and Yang, An and Du, Sean and Zheng, Chujie and Huang, Fei and Liu, Dayiheng and Huang, Gao and Zhou, Jingren},
  journal = {Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing},
  year = {2026},
  month = {Jan},
}