superscript

The passage around this formula

Shazeer and colleagues stated the underlying argument for conditional computation directly: “a trainable gating network determines a sparse combination of experts to use for each example,” a mechanism they showed could scale model capacity by “over 1000x” while keeping the compute spent on any one example roughly constant [ 11 ] . The now-standard form of a sparse mixture-of-experts layer routes each token to a small top- k subset of E available experts: $y(x) = \sum_{i \in \mathrm{TopK}(G(x))} G(x)_i \cdot E_i(x), \qquad C_{\text{tok}} \approx \frac{k}{E} \cdot C_{\text{tok}}^{\text{dense}}(N_{\text{total}})$. with G(x) a learned gating distribution over experts. Compute per token scales with the active fraction k/E , not with the total parameter count $N_{\text{total}}$ — the whole strategy in one line.

Read this part in the article →