Skip to content
Get in touch

Reviewed in POC · content version 2

Review Parameters

Your draft is bound to this exact source hash. If the term changes, the Review Desk will use a new storage key instead of silently attaching old decisions.

Definition

Parameters are the adjustable numerical values a model uses to transform inputs into outputs; many are learned during training.
Decision for Definition

Plain-Language Explanation

Training changes model parameters so the system performs better according to its training objective. Parameter count describes one aspect of model size, not the quality of the whole system.
Decision for Plain-Language Explanation

Why It Matters

Parameter counts are often used in product comparisons, but architecture, data, training, tools, and evaluation can matter as much or more for a specific task.
Decision for Why It Matters

Example

A model described as having billions of parameters contains billions of adjustable values distributed across its learned structure.
Decision for Example

Common Misconception

More parameters do not automatically mean a model is more accurate, safer, more efficient, or better suited to a particular use.
Decision for Common Misconception

Remember It This Way

Parameters are the model's adjustable internal values.
Decision for Remember It This Way

Related Terms

* [[weights|Weights]] * [[neural-network|Neural Network]] * [[training-data|Training Data]] * [[model-distillation|Model Distillation]]
Decision for Related Terms

Sources

* [Google for Developers, Machine Learning Glossary](https://developers.google.com/machine-learning/glossary) — technical definition of model parameters and related training concepts.
Decision for Sources

Learn More

* Relate parameter count to [[weights|Weights]] and efficiency work such as [[model-distillation|Model Distillation]].
Decision for Learn More
Return to term

Not saved yet.