Transformer Dynamics: A neuroscientific approach to interpretability of large language models

TL;DR

This paper introduces a neuroscientific-inspired dynamical systems framework to interpret the internal mechanisms of transformer models, revealing stable and unstable dynamics in the residual stream across layers.

Contribution

It pioneers a dynamical systems approach to analyze transformer residual streams, bridging neuroscience and AI interpretability with novel insights into layer-wise activations.

Findings

Residual stream units show continuity across layers

Activations accelerate and become denser with depth

RS exhibits attractor-like dynamics in lower layers

Abstract

As artificial intelligence models have exploded in scale and capability, understanding of their internal mechanisms remains a critical challenge. Inspired by the success of dynamical systems approaches in neuroscience, here we propose a novel framework for studying computations in deep learning systems. We focus on the residual stream (RS) in transformer models, conceptualizing it as a dynamical system evolving across layers. We find that activations of individual RS units exhibit strong continuity across layers, despite the RS being a non-privileged basis. Activations in the RS accelerate and grow denser over layers, while individual units trace unstable periodic orbits. In reduced-dimensional spaces, the RS follows a curved trajectory with attractor-like dynamics in the lower layers. These insights bridge dynamical systems theory and mechanistic interpretability, establishing a foundation for a "neuroscience of AI" that combines theoretical rigor with large-scale data analysis to advance our understanding of modern neural networks.