Search papers, labs, and topics across Lattice.
The paper analyzes the superior scaling properties of Gated Linear Units (GLUs) compared to MLPs using numerical analysis, attributing the difference to the piecewise quadratic functional form of GLUs enabling quadratic order approximation. It demonstrates that the scaling slope L(P) scales as P^{-3} for GLUs, faster than the P^{-2} scaling for MLPs in function reconstruction. The authors then propose a "Gated Quadratic Unit" with an even steeper scaling slope, suggesting a path towards architecture design guided by numerical theory.
GLUs' secret weapon is out: their piecewise quadratic nature allows them to scale asymptotically faster than MLPs (L(P) ∝ P⁻³ vs. P⁻²), potentially revolutionizing architecture design.
Scaling laws can be understood from ground-up numerical analysis, where traditional function approximation theory can explain shifts in model architecture choices. GLU variants now dominate frontier LLMs and similar outer-product architectures are prevalent in ranking models. The success of these architectures has mostly been left as an empirical discovery. In this paper, we apply the tools of numerical analysis to expose a key factor: these models have an $x^2$ which enables \emph{asymptotically} faster scaling than MLPs. GLUs have piecewise quadratic functional forms that are sufficient to exhibit quadratic order of approximation. Our key contribution is to demonstrate that the $L(P)$ scaling slope is $L(P)\propto P^{-3}$ for GLUs but only $L(P)=P^{-2}$ for MLPs on function reconstruction problems. We provide a parameter construction and empirical verification of these slopes for 1D function approximation. From the first principles we discover, we make one stride and propose the ``Gated Quadratic Unit'' which has an even steeper $L(P)$ slope than the GLU and MLP. This opens the possibility of architecture design from first principles numerical theory to unlock superior scaling in large models. Replication code is available at https://github.com/afqueiruga/divine_scaling.
