arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.08088cs.LGcs.AI
收藏

在线领域感知的LLM解码用于持续领域演变

Online Domain-aware LLM Decoding for Continual Domain Evolution

University of North Texas(北卡罗来纳州立大学)

Mohammad Abu-Shaira, Weishi Shi

AI总结:

本文提出在线领域感知解码框架ODD,通过概率融合和自适应置信度调节,提升LLM在持续领域变化中的适应能力,实验表明其在语法和语义生成任务中表现优异。

摘要:

LLMs通常在领域特定数据上离线微调,假设领域静态。但实际上,领域知识通过新法规、产品、服务和交互模式持续演变。对每个新实例重新训练或微调LLM在计算上不可行。此外,现实环境也表现出时间动态性,数据分布不断变化。忽视这种现象,即概念漂移,会显著降低模型的预测准确性。这种领域演变与静态适应管道的不匹配凸显了需要高效实时适应而无需昂贵再训练的需求。为此,我们引入在线领域感知解码框架(ODD)。ODD在基础LLM和前缀树先验之间进行概率级融合,通过自适应置信度调节使用分歧和连续性信号进行指导。在多样化的漂移场景下的实证评估表明,ODD在所有语法和语义NLG指标上均优于LLM-Greedy和LLM-Temp Scaled。它在ROUGE-L指标上获得绝对增益0.065,并在最佳基线上使余弦相似度提高13.6%。这些结果证明了ODD对演变词汇和上下文模式的鲁棒性,使其适用于动态LLM应用。

英文摘要:

LLMs are typically fine-tuned offline on domain-specific data, assuming a static domain. In practice, domain knowledge evolves continuously through new regulations, products, services, and interaction patterns. Retraining or fine-tuning LLMs for every new instance is computationally infeasible. Additionally, real-world environments also exhibit temporal dynamics with shifting data distributions. Disregarding this phenomenon, commonly referred to as concept drift, can significantly diminish a model's predictive accuracy. This mismatch between evolving domains and static adaptation pipelines highlights the need for efficient, real-time adaptation without costly retraining. In response, we introduce Online Domain-aware Decoding framework (ODD). ODD performs probability-level fusion between a base LLM and a prefix-tree prior, guided by adaptive confidence modulation using disagreement and continuity signals. Empirical evaluation under diverse drift scenarios demonstrates that ODD consistently surpasses LLM-Greedy and LLM-Temp Scaled across all syntactic and semantic NLG metrics. It yields an absolute ROUGE-L gain of 0.065 and a 13.6% relative improvement in Cosine Similarity over the best baseline. These results demonstrate ODD 's robustness to evolving lexical and contextual patterns, making it suitable for dynamic LLM applications.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑