Category: File Formats
Definition
CKPT (checkpoint) files save the complete state of a neural network during training, including weights, optimizer state, and training progress.
How It Works
Checkpoints capture snapshots of model parameters at specific training intervals. They enable resuming interrupted training and comparing model performance at different stages.
The format typically uses Python's pickle protocol, making it framework-specific but comprehensive.
Why It Matters
Checkpoint files are essential for long training runs that might be interrupted. They allow researchers to save progress, experiment with different training paths, and share pre-trained models.
They're the primary format for distributing Stable Diffusion and other generative models.
← Back to File Formats | All Terms