Skip to main content
All terms
Foundations

Model Checkpoint

A saved snapshot of a model's state taken during or after training.

Definition

A model checkpoint is a saved snapshot of a model's state, usually its weights and sometimes its optimizer state, captured at a point during or after training. Checkpoints let a run resume after interruption, allow comparison of versions, and provide the artifact that is loaded for inference. They are how training progress is preserved and shared.