总结事实：LLM中事实召回背后的加法机制
Summing Up the Facts: Additive Mechanisms Behind Factual Recall in LLMs
AI总结：
研究LLM事实召回机制，发现其由多个独立机制加法组合产生相长干涉，扩展直接logit归因方法揭示混合头结构。
摘要：
基于Transformer的大型语言模型（LLM）如何存储和检索知识？我们关注该任务最基本的形式——事实召回，即模型被要求在形如“Fact: The Colosseum is in the country of”的提示中显式呈现存储的事实。我们发现事实召回背后的机制故事比先前认为的更为复杂。它包含几个独立且性质不同的机制，这些机制以加法方式结合，在正确属性上产生相长干涉。我们将这种普遍现象称为加法模式：模型通过累加多个独立贡献进行计算。每个机制单独的贡献可能不足，但累加会在正确答案上产生相长干涉。此外，我们将直接logit归因方法扩展为将注意力头的输出归因于单个源token。我们利用该技术解构了所谓的“混合头”——它们本身是来自不同源token的两次独立加法更新的组合。
英文摘要：
How do transformer-based large language models (LLMs) store and retrieve knowledge? We focus on the most basic form of this task -- factual recall, where the model is tasked with explicitly surfacing stored facts in prompts of form `Fact: The Colosseum is in the country of'. We find that the mechanistic story behind factual recall is more complex than previously thought. It comprises several distinct, independent, and qualitatively different mechanisms that additively combine, constructively interfering on the correct attribute. We term this generic phenomena the additive motif: models compute through summing up multiple independent contributions. Each mechanism's contribution may be insufficient alone, but summing results in constructive interfere on the correct answer. In addition, we extend the method of direct logit attribution to attribute an attention head's output to individual source tokens. We use this technique to unpack what we call `mixed heads' -- which are themselves a pair of two separate additive updates from different source tokens.
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。
