Review Inference

Inference is the process of running a trained model on an input to produce a prediction, classification, generation, or other output.

Training is when the model's learned values are adjusted. Inference is when the trained model is used, such as when it answers a prompt or analyzes a new image.

Inference determines run-time speed, cost, hardware needs, privacy exposure, and how a model is integrated into a real workflow.

When a user submits a prompt and the model generates a response, the model is performing inference.

Ordinary inference does not necessarily update the model's weights or make it learn permanently from each request.

Training changes the model; inference uses it.

* [[machine-learning|Machine Learning]] * [[training-data|Training Data]] * [[latency|Latency]] * [[compute|Compute]] * [[input|Input]] * [[output|Output]]

* [Google for Developers, *Machine Learning Glossary: Inference*](https://developers.google.com/machine-learning/glossary#inference) — technical definition of applying a trained model to unlabeled examples to make predictions. * [NIST, *Adversarial Machine Learning: A Taxonomy and Terminology of Attacks and Mitigations* (NIST AI 100-2 E2025)](https://doi.org/10.6028/NIST.AI.100-2e2025) — lifecycle terminology distinguishing training and inference stages.

* Connect run-time behavior to [[latency|Latency]] and [[compute|Compute]].