Open Models Almanac
Methodsinference & servingreasoning control

single mention · inference & serving

Capped linear reasoning-token length deduction

A linear deduction on reasoning-token length is applied with a cap, as distinct from the effort-dependent exponential penalty schedule.

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

Also called length deduction.

Filed under inference & serving › reasoning control