We have proposed that this function [of sleep] is to renormalize synaptic weights after learning has led to a net increase in synaptic strength in many brain circuits. Without this renormalization, synaptic activity would become energetically too expensive and saturation would prevent new learning. There is converging evidence from molecular, electrophysiological, and ultrastructural experiments showing a net increase in synaptic strength after the major wake phase, and a net decline after sleep.

https://www.sciencedirect.com/science/article/abs/pii/S1084952121000318