Two-Clustering Regime of Token Dynamics in Causal Attention
Organizations: School of Mathematical Sciences Shanghai Jiao Tong University, China · Khoury College of Computer Sciences Northeastern University, USA · Institute of Mathematics Vietnam Academy of Science and Technology · University of Science and Technology of Hanoi Vietnam Academy of Science and Technology ·
Abstract
We partially resolve Conjecture 1 of [Karagodin et al., Clustering in Causal Attention Masking (NeurIPS 2024)] and extend it to time-dependent parameter regimes. The conjecture predicts that causal attention dynamics converge to two clusters aligned with the eigenvectors associated with the largest eigenvalue of the value transformation. We prove that this behavior holds when and is generally false when . The resulting two-cluster regime is common in practice and substantially more challenging than the classical single-cluster setting studied previously. A central contribution of our work is the identification of a rigorous dimension reduction phenomenon: despite evolving in a high-dimensional embedding space, token trajectories collapse onto a two-dimensional structure governed by a small number of dominant directions. Building on this insight, we introduce a new analytical framework--instability within the flow--inspired by instability phenomena in physical systems. This framework enables the analysis of steady-regime behavior in non-monotone, non-autonomous attention dynamics. Our theoretical findings are illustrated through simulations and experimental studies using Group Query Attention (GQA) layers.
Figures & tables
| Work | Parameters | Initial Data | Key Techniques | Theoretical Results |
|---|---|---|---|---|
| Karagodin et al. (2024) | constant | Arbitrary | Explicit formula; | Convergence of to |
| general | on | Schub genericity theorem | Full convergence if | |
| Abella et al. (2025) | symmetric | Single | Lyapunov function; | Global convergence to |
| general | hemisphere | Input-to-state stability | single cluster | |
| Ours | symmetric | Arbitrary | Instability analysis; | Global convergence to |
| general | on | Hierarchy dim. deduction | two clusters ; |
| Fine-tuning weight | |||||
|---|---|---|---|---|---|
| Fine-tuning rank | 8 | 8 | 8 | 4 | 4 |
| WikiSQL ( ) | 70.4 | 70.0 | 73.0 | 71.4 | 73.7 |
| MultiNLI ( ) | 91.0 | 90.8 | 91.0 | 91.3 | 91.3 |