arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2504.01002cs.CLcs.AI
收藏

Token嵌入违反流形假设

Token embeddings violate the manifold hypothesis

Mathematics and Statistics(数学与统计学);American University(美国大学);Galois, Inc.(Galois公司);Department of Mathematics, University of Washington(华盛顿大学数学系)

Michael Robinson, Sourya Dey, Tony Chiang

AI总结:

本文通过提出纤维丛假设检验,从经验和理论上证明LLM的词元嵌入空间不满足流形假设,并发现涉及不规则词元的提示响应稳定性更差。

摘要:

全面理解大型语言模型(LLM)的行为需要掌握其输入词元空间。如果该空间与我们的假设不同,我们对LLM的理解和结论很可能是错误的。我们从经验和理论上阐明了词元嵌入的结构。我们提出了一种新颖的统计检验,以每个词元周围的邻域具有相对平坦且光滑的结构作为零假设。未能拒绝零假设是无信息的,但在特定词元ψ处拒绝它意味着在ψ邻域B(ψ)内的词元子空间存在不规则性。零假设中假设的结构是带边界流形的一种推广,称为光滑纤维丛(可分解为两个空间区域——小半径和大半径),因此我们将新的假设检验称为“纤维丛假设”。通过在多个具有独特词元嵌入的开源LLM上运行我们的检验,我们发现零假设经常被拒绝,因此证据表明词元子空间不是纤维丛,因而也不是流形。作为我们发现的后果,当LLM被呈现两个语义等价的提示时,如果其中一个提示包含被我们检验所涉及的词元,那么该提示的响应可能比另一个表现出更少的不稳定性。

英文摘要:

A full understanding of the behavior of a large language model (LLM) requires our grasp of its input token space. If this space differs from our assumptions, our comprehension of and conclusions about the LLM will likely be flawed. We elucidate the structure of the token embeddings both empirically and theoretically. We present a novel statistical test assuming that the neighborhood around each token has a relatively flat and smooth structure as the null hypothesis. Failing to reject the null is uninformative, but rejecting it at a specific token $ψ$ implies an irregularity in the token subspace in a $ψ$-neighborhood, $B(ψ)$. The structure assumed in the null is a generalization of a manifold with boundary called a \emph{smooth fiber bundle} (which can be split into two spatial regimes -- small and large radius), so we denote our new hypothesis test as the ``fiber bundle hypothesis.'' By running our test over several open-source LLMs, each with unique token embeddings, we find that the null is frequently rejected, and so the evidence suggests that the token subspace is not a fiber bundle and hence also not a manifold. As a consequence of our findings, when an LLM is presented with two semantically equivalent prompts, if one prompt contains a token implicated by our test, the response to that prompt will likely exhibit less stability than the other.

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。

↑