Open Models Almanac
Methodsoptimizationtraining parallelism

single source · optimization

MoonEP

MoonEP balances expert execution using dynamic redundant experts and online planning and migration; the evidence also describes minimizing maximum redundant experts per rank and static computation shapes with zero-copy communication.

The sources on this page name it without explaining how it works. The line above is the extraction's one-line gloss. See training parallelism for the methods filed beside it.

Also called MoonEP expert placement planning.

Filed under optimization › training parallelism