test

Definition

Evaluation, almost always inside a compound. A test set is data held out from training; test-time means after training, at the moment of inference. The second sense drives a major current line of work — test-time compute, where accuracy is bought by thinking longer, sampling multiple attempts, or searching at inference rather than by making the model bigger. Training compute and test-time compute are separate budgets.

Explainers using this term

• TTPO Explained: Training a Model Mid-Exam, With No Answer KeyTTPO: Test-Time Policy Optimization