single mention · optimization
Prefill context parallelism
This method splits each prompt across ranks during prefill.
The sources on this page only mention it in passing. The line above is the extraction's one-line gloss. See training parallelism for the methods filed beside it.
Filed under optimization › training parallelism