Open Models Almanac
Methodsoptimizationtraining stability

single mention · optimization

Training-only embedding input dropout

Applies 10% dropout to embeddings during training while omitting internal dropout layers.

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