Open Models Almanac
Methodsoptimizationtraining stability

single mention · optimization

Weight decay coupled to learning rate squared

Sets the weight-decay strength proportional to the square of the learning rate to help maintain model-weight size across training horizons.

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

Filed under optimization › training stability