真相留在家族中：通过模型谱系中继承的真相头增强上下文基础
The Truth Stays in the Family: Enhancing Contextual Grounding via Inherited Truthful Heads in Model Lineages
AI总结：
研究发现基础LLM与下游变体间存在上下文真相分数的强继承性，提出TruthProbe软门控策略放大真相头以提升上下文真实性并减少多模态幻觉。
摘要：
大型语言模型（LLM）的最新进展产生了许多共享基础LLM的专业多模态LLM（MLLM），形成了不同的模型谱系。基础LLM与下游变体之间是否存在基本的行为联系尚不清楚。我们通过量化头部级别的上下文真相分数来研究这个问题。在包括基于Vicuna、Qwen2.5、LLaMA2和Mistral的模型在内的多种LLM和MLLM谱系中，我们发现真相分数在模型家族内被强烈保留，即使在指令调优或多模态适应后也是如此。我们进一步表明，这种继承与注意力头权重保留一致，并且上下文真相头关注查询相关的证据。基于这一发现，我们提出了TruthProbe，一种软门控策略，在保留其他头部贡献的同时放大上下文真相头。TruthProbe在HaluEval上提高了上下文真实性，并在POPE和CHAIR上减少了多模态幻觉，基础LLM的真相分数有效转移到其微调的LLM和MLLM后代。代码可在https://github.com/miso-choi/TruthProbe获取。
英文摘要：
Recent advances in large language models (LLMs) have produced many specialized multimodal LLMs (MLLMs) that share common foundational LLMs, forming distinct model lineages. It remains unclear whether a fundamental behavioral link exists between the foundational LLMs and downstream variants. We investigate this question by quantifying head-level context-truthfulness scores. Across diverse LLM and MLLM lineages, including Vicuna-, Qwen2.5-, LLaMA2-, and Mistral-based models, we find that Truth Scores are strongly preserved within model families, even after instruction tuning or multimodal adaptation. We further show that this inheritance is consistent with attention-head weight preservation, and that context-truthful heads attend to query-relevant evidence. Building on this finding, we propose TruthProbe, a soft-gating strategy that amplifies context-truthful heads while preserving other head contributions. TruthProbe improves contextual truthfulness on HaluEval and reduces multimodal hallucination on POPE and CHAIR, with base-LLM Truth Scores transferring effectively to their fine-tuned LLM and MLLM descendants. Code is available at https://github.com/miso-choi/TruthProbe.
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。
