研究人员开发了一种名为 Logit-Contribution Scoring (LOCOS) 的新方法，用于识别大语言模型中的非字面检索头。与之前关注字面词元匹配的方法不同，LOCOS 分析注意力头的输出值电路，以了解它们如何从上下文中综合信息。这种方法在检测负责非字面检索的头方面显示出更大的有效性，涵盖了 Qwen3、Gemma-3 和 OLMo-3.1 等各种模型系列，当这些已识别的头被消融时，会导致需要综合的任务性能显著下降。 AI
影响 提供了一种更准确的方法来解释大语言模型如何综合信息，这对于理解和改进长上下文能力至关重要。
排序理由 介绍分析大语言模型行为新方法的学术论文。[lever_c_demoted from research: ic=1 ai=1.0]
AI 生成摘要 · Google Gemini · 来自 2 个来源。 我们如何撰写摘要 →
arXiv:2607.01002v1 Announce Type: cross Abstract: In long-context use, large language models frequently synthesize answers from the meaning of a relevant context span rather than literally copy-pasting them. Identifying which attention heads perform this synthesis matters for int…
In long-context use, large language models frequently synthesize answers from the meaning of a relevant context span rather than literally copy-pasting them. Identifying which attention heads perform this synthesis matters for interpreting long-context model behavior. Yet existin…