On the Geometric Structure of Layer Updates in Deep Language Models

TL;DR

This paper investigates the geometric structure of layer updates in deep language models, revealing a dominant tokenwise component and a distinct residual that influences model behavior.

Contribution

It introduces a framework to analyze layer update geometry, showing the residual's distinctness and its functional impact across various architectures.

Findings

Layer updates are mostly aligned with a tokenwise component.

The residual component is geometrically distinct and less aligned.

Approximation error correlates strongly with output perturbation.

Abstract

We study the geometric structure of layer updates in deep language models. Rather than analyzing what information is encoded in intermediate representations, we ask how representations change from one layer to the next. We show that layerwise updates admit a decomposition into a dominant tokenwise component and a residual that is not captured by restricted tokenwise function classes. Across multiple architectures, including Transformers and state-space models, we find that the full layer update is almost perfectly aligned with the tokenwise component, while the residual exhibits substantially weaker alignment, larger angular deviation, and significantly lower projection onto the dominant tokenwise subspace. This indicates that the residual is not merely a small correction, but a geometrically distinct component of the transformation. This geometric separation has functional consequences: approximation error under the restricted tokenwise model is strongly associated with output perturbation, with Spearman correlations often exceeding 0.7 and reaching up to 0.95 in larger models. Together, these results suggest that most layerwise updates behave like structured reparameterizations along a dominant direction, while functionally significant computation is concentrated in a geometrically distinct residual component. Our framework provides a simple, architecture-agnostic method for probing the geometric and functional structure of layer updates in modern language models.