Training
Early stopping
What is early stopping?
Early stopping ends a training run when validation performance stops improving, rather than running to the planned number of steps. It prevents overfitting and stops paying for compute that is making the model worse.
It requires a validation split evaluated periodically. The usual rule is patience-based: stop if validation loss has not improved for a set number of evaluations, and keep the best checkpoint rather than the last.
On short fine-tuning runs the saving is often more about quality than money — the run was going to be cheap anyway, but the last thousand steps were actively harmful.
Related terms
Start with the free tier
A magic link creates your account, your tenant and your first API key. No card until you ask for compute.