Open Models Almanac
Methodsdata curationdata mixture & curriculum

single source · data curation

SFT sampling proportional to unique-token counts

Sampling datasets in proportion to their unique-token counts, including as a rule for within-domain shares.

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.

Also called natural rule.

Filed under data curation › data mixture & curriculum