Open Models Almanac
Methodspost-training

single mention · post-training

Dataset-specific reference groups for reasoning-effort labeling

A labeling scheme that assigns reasoning-effort labels by fitting thresholds within reference groups rather than globally.1 Each group is a (dataset, difficulty) pair, so prompts are compared against peers of the same task format and difficulty instead of a global reasoning-length threshold.2

How this definition was written

Summarised by glm-5.3-flash from 1 source; every sentence is checked against the quote it cites.

  1. we adopt dataset-specific reference groups — tech report
  2. We therefore fit the thresholds per reference group g = (dataset, difficulty — tech report

Also called we adopt dataset-specific reference groups.

Filed under post-training