Open Models Almanac
Methodsoptimizationtraining runtime

single mention · optimization

Asymmetric local-read and single-writer cache paths

Uses local cache reads and a designated sidecar to archive caches to shared storage, avoiding simultaneous writes by all nodes.

The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See training runtime for the methods filed beside it.

Also called Asymmetric read/write cache paths.

Filed under optimization › training runtime