Open Models Almanac
Methodsoptimizationtraining runtime

single mention · optimization

Recompute MoE feed-forward activations instead of checkpointing them

Avoids checkpointing activations inside the MoE feed-forward network to reduce device memory use.

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

Filed under optimization › training runtime