Large language models (LLMs) often exhibit Context Faithfulness Hallucinations, where outputs deviate from retrieved information due to incomplete context integration. Our analysis reveals a strong correlation between token-level uncertainty and hallucinations. We hypothesize that attention mechanisms inherently encode context utilization signals, supported by probing analysis. Based on these insights, we propose Dynamic Attention-Guided Context Decoding (DAGCD), a lightweight framework that leverages attention distributions and uncertainty signals in a single-pass decoding. Experiments on open-book QA datasets demonstrate DAGCD's effectiveness, yielding significant improvements in faithfulness and robustness while preserving computational efficiency.
@article{arxiv.2501.01059,
title = {Dynamic Attention-Guided Context Decoding for Mitigating Context Faithfulness Hallucinations in Large Language Models},
author = {Yanwen Huang and Yong Zhang and Ning Cheng and Zhitao Li and Shaojun Wang and Jing Xiao},
journal= {arXiv preprint arXiv:2501.01059},
year = {2025}
}