Quantifying Concentration Phenomena of Mean-Field Transformers in the Low-Temperature Regime

arXiv:2605.10931v1 Announce Type: cross Abstract: Transformers with self-attention modules as their core components have become an integral architecture in modern large language and foundation models. In this paper, we study the evolution of tokens in deep encoder-only transformers at inference time which is described in the large-token limit by a mean-field continuity equation. Leveraging ideas from the convergence analysis of interacting multi-particle systems, with particles corresponding to tokens, we prove that the token distribution rapidly concentrates onto the push-forward of the initial distribution under a projection map induced by the key, query, and value matrices, and remains metastable for moderate times. Specifically, we show that the Wasserstein distance of the two distributions scales like sqrt{{log(eta+1)}/{eta}}exp(Ct)+exp(-ct) in terms of the temperature parameter eta^{-1}o 0 and inference time tgeq 0. For the proof, we establish Lyapunov-type estimates for the zero-temperature equation, identify its limit as toinfty, and employ a stability estimate in Wasserstein space together with a quantitative Laplace principle to couple the two equations. Our result implies that for time scales of order logeta the token distribution concentrates at the identified limiting distribution. Numerical experiments confirm this and, beyond that, complement our theory by showing that for finite eta and large t the dynamics enter a different terminal phase, dominated by the spectrum of the value matrix.

Source: arXiv cs.LG | 2026-05-12