checkpointing

Periodic state saving

Checkpointing is the practice of saving the state of a system or process at intervals so work can be resumed from that point if something fails.

In computing, long-running programs use checkpointing to avoid losing hours of progress to a crash: The model training job checkpoints every thousand steps.

Video games use the same principle with save points.

computing

snapshot

Informal photograph taken quickly

commit · state · persistence · recovery · serialization · journaling · transaction · versioning · backup · failover

gaming

autosave

Automatic file preservation

progress · bossfight · mission · quest · level · inventory · gameplay · continue · restart · savefile

machine-learning

epoch

Distinct historical period

  • trainingDeveloping abilities through practice
  • modelRepresentation or paradigm
  • weightsHeavy objects for lifting

gradient · iteration · resumption · tensor · validation · loss · optimizer · parameters · batch · convergence

Categories
Opposites
data persistence strategy

checkpointingjournaling

Regular personal writing practice

recovery approach

checkpointingcold start

Starting from zero

crash and burn · fire-and-forget

Family

check, checked, checking, checkers, checker, checklist, checkpoint, double-check, checklists

Familiarity
Top 500,000
Profile
Used worldwide · Timeless · Rated G