Open Models Almanac
Methodsoptimizationtraining parallelism

single mention · optimization

Latency-optimized context-parallel collectives

Context-parallel collectives use latency-optimized exchange patterns, in-kernel compute-communication overlap, and fusion.

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