YARD: Y-Architecture Register Decoding for Efficient Hallucination Mitigation in Large Vision-Language Models

Ting Chen, Geng Li, Guohao Chen, Yu Hu, Guan Huang, Mai Chen, Langsheng Lei, Jun Du · 2026

Read YARD: Y-Architecture Register Decoding for Efficient Hallucination Mitigation in Large Vision-Language Models by Ting Chen, Geng Li, Guohao Chen, Yu Hu, Guan Huang, Mai Chen, Langsheng Lei, Jun Du (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-31429.txt.

Description

Chen and colleagues propose YARD, a training-free contrastive-decoding framework to cut hallucinations in large vision-language models. Existing methods degrade badly: dropping visual tokens induces language hallucinations, while corrupting images needs two passes. YARD builds the degraded branch internally, sharing shallow computation and branching at middle layers. It swaps patch visual tokens for register tokens keeping global meaning without fine local evidence. That gives a faithful contrast signal without a second forward pass. Tests show state-of-the-art mitigation across several models, with much lower latency.

Catalog record

CREATORTing Chen, Geng Li, Guohao Chen, Yu Hu, Guan Huang, Mai Chen, Langsheng Lei, Jun Du
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE72638
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2605.31429

Browse the full library · About the Tower of Babel