A checkpoint is a saved snapshot of a model's learned values, so training can stop and pick up later, or the model can be loaded and used.
In plain words
Training a model can take hours or days. Saving checkpoints along the way means a crash or a planned stop does not lose the work: you reload the latest one and continue, even on a different machine. To resume training properly you save more than the model itself, such as the optimizer's state.
A quick example
Imagine a training run that saves a checkpoint every hour. On the second day the machine restarts. The team loads the last checkpoint on a new machine and training continues from that hour instead of starting over.
Where you'll meet it
- Long training runs that may be interrupted.
- Downloading a trained model to use or fine-tune.
Related
Sources
- Docscheckpoint, Google Machine Learning Glossary · read 29 Sept 2026
- DocsSaving and Loading Models, PyTorch · read 29 Sept 2026
Checked by WhatIsAI against the sources above · 29 Sept 2026. Spotted an error? Report it.