Open Models Almanac
Methodsoptimizationtraining parallelism

single mention · optimization

All-to-all gradient exchange with local FP32 summation

This gradient-reduction procedure exchanges local gradients across ranks by all-to-all and then sums locally in FP32.

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