arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2501.07359cs.CLcs.AI
收藏

规模扩展对大语言模型内部功能层级的涌现效应

Emergent effects of scaling on the functional hierarchies within large language models

Duke University(杜克大学)

Paul C. Bogdan

AI总结:

本研究探讨规模扩展对大语言模型功能层级的影响,发现小模型部分符合层级表征规律,而大模型存在抽象层级波动、相邻层注意力协同等偏离稳定层级的涌现效应。

摘要:

大语言模型(LLM)架构常被描述为具有功能层级性:早期层处理句法,中间层开始解析语义,晚期层整合信息。本研究重新探讨了这些观点。研究向LLM输入简单文本(例如“A church and organ”)并提取由此产生的激活值。随后,针对每一层,研究拟合支持向量机和岭回归模型来预测文本标签,以此检验某一特定层是否编码了某种信息。使用小型模型(Llama-3.2-3b;28层)的分析部分支持了普遍的层级观点:项目级语义在早期(第2-7层)表征最强,随后是双项目关系(第8-12层),再之后是四项目类比(第10-15层)。此后,项目和简单关系的表征在更专注于全局信息的深层中逐渐减弱。然而,若干发现与稳定层级的观点相悖:第一,尽管深层能够表征文档级别的抽象概念,但深层也会压缩上下文窗口早期部分的信息,且未进行有意义的抽象。第二,在研究更大的模型(Llama-3.3-70b-Instruct)时,抽象层级出现了剧烈波动:随着深度增加,双项目关系和四项目类比的表征强度先上升,随后显著下降,之后又短暂回升。这种特殊模式在多项实验中一致出现。第三,规模扩展的另一个涌现效应是相邻层注意力机制之间的协同性。在使用该更大模型的多项实验中,相邻层各自专门表征的信息会发生波动。总而言之,抽象层级通常会在各层之间显现,但大模型也会以奇特的方式偏离这种结构。

英文摘要:

Large language model (LLM) architectures are often described as functionally hierarchical: Early layers process syntax, middle layers begin to parse semantics, and late layers integrate information. The present work revisits these ideas. This research submits simple texts to an LLM (e.g., "A church and organ") and extracts the resulting activations. Then, for each layer, support vector machines and ridge regressions are fit to predict a text's label and thus examine whether a given layer encodes some information. Analyses using a small model (Llama-3.2-3b; 28 layers) partly bolster the common hierarchical perspective: Item-level semantics are most strongly represented early (layers 2-7), then two-item relations (layers 8-12), and then four-item analogies (layers 10-15). Afterward, the representation of items and simple relations gradually decreases in deeper layers that focus on more global information. However, several findings run counter to a steady hierarchy view: First, although deep layers can represent document-wide abstractions, deep layers also compress information from early portions of the context window without meaningful abstraction. Second, when examining a larger model (Llama-3.3-70b-Instruct), stark fluctuations in abstraction level appear: As depth increases, two-item relations and four-item analogies initially increase in their representation, then markedly decrease, and afterward increase again momentarily. This peculiar pattern consistently emerges across several experiments. Third, another emergent effect of scaling is coordination between the attention mechanisms of adjacent layers. Across multiple experiments using the larger model, adjacent layers fluctuate between what information they each specialize in representing. In sum, an abstraction hierarchy often manifests across layers, but large models also deviate from this structure in curious ways.

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。

↑