Open Models Almanac
Methodsoptimizationtraining runtime

single mention · optimization

Policy-model gradient-buffer reuse for reference-model weights

Stores reference-model weights in CPU memory and materializes them using the policy model's gradient-buffer storage when needed.

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