Open Models Almanac
Methodsdata curationdata mixture & curriculum

single mention · data curation

Log-linear benchmark models for optimizing mid-training domain weights

Fitting per-benchmark log-linear laws to expert-merge scores to select domain weights maximizing aggregate benchmark performance.

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

Filed under data curation › data mixture & curriculum