Open Models Almanac
Methodsoptimizationtraining precision

single mention · optimization

Mixed-precision policy with BF16 storage and communication and FP32 reductions

A precision policy using BF16 for storage and communication while performing reductions in FP32.

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

Also called mixed-precision policy.

Filed under optimization › training precision