Zubnet AILearnWiki › Early Stopping
Training

Early Stopping

Patience, Validation-Based Stopping
Stopping training when performance on a held-out validation set stops improving, rather than training for a fixed number of steps. As training continues, training loss keeps decreasing but validation loss eventually starts increasing — the model is overfitting to training data. Early stopping catches this inflection point and saves the best model before quality degrades.

Why it matters

Early stopping is the simplest and most effective regularization technique for fine-tuning. Without it, you risk training too long and destroying the capabilities you wanted to preserve. With it, the model automatically stops at its best point. The "patience" parameter (how many evaluations without improvement before stopping) is one of the most important hyperparameters in fine-tuning.

Deep Dive

The process: (1) split your data into training and validation sets, (2) evaluate on the validation set periodically during training, (3) track the best validation metric (loss, accuracy, F1), (4) if the metric hasn't improved for N evaluations (patience), stop training and revert to the checkpoint with the best validation score. This prevents the model from memorizing training data beyond the point where it helps generalization.

In LLM Fine-Tuning

For LLM fine-tuning, early stopping is especially important because catastrophic forgetting can destroy base model capabilities. A model fine-tuned for too long on customer support data might become great at support but lose its ability to do math or write code. Monitoring validation loss across multiple task types (not just the fine-tuning task) helps catch this. Typical fine-tuning runs are 1–5 epochs with patience of 2–3 evaluations.

Not Used in Pre-Training

Interestingly, LLM pre-training rarely uses early stopping. The training runs are so expensive and the datasets so large that models typically train for a predetermined number of tokens (based on scaling laws). Overfitting is rare during pre-training because the model usually never sees the same data twice. Early stopping is primarily a fine-tuning and classical ML technique.

Related Concepts

← All Terms
ESC