Open Models Almanac
Methodsoptimizationtraining runtime

single mention · optimization

Slice-Granularity Elasticity

Slice-Granularity Elasticity is a training-runtime method that lets training continue on fewer slices of TPU chips after a localized failure.1 It reconfigures the hardware so interruptions cause only seconds of delay instead of many minutes.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. allows continuous training with fewer “slices” of TPU chips when there is a localized failure — Gemma 4 Technical Report
  2. This reconfiguration reduces the delay caused by interruptions from many minutes to a few seconds — Gemma 4 Technical Report

Filed under optimization › training runtime