Model Training · Level 2 of 5
Checkpoint
A saved snapshot of training state or model parameters.
Resuming exactly may also require optimizer, scheduler, and random-generator state.
Listen to the definition and example
Audio transcript
Checkpoint. A saved snapshot of training state or model parameters. Resuming exactly may also require optimizer, scheduler, and random-generator state. For example: Training resumes from saved weights and optimizer moments after a restart.
Explore this concept
Start with
Related concepts
Quick recall question
Try answering before looking back at the definition.