Open Models Almanac
Methodsoptimizationtraining parallelism

single mention · optimization

Gradient-hook synchronization for replicated LoRA parameters

Gradient hooks synchronize unsharded replicated LoRA parameter gradients across ranks before the optimizer step.

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