Open Models Almanac
Methodsoptimizationtraining runtime

single mention · optimization

Double-buffered chunked streaming of reference-model weights

Streams reference-model weights in chunks, using one slot for computation while another prefetches the next chunk.

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