parameter

A parameter is a learned internal value, such as a weight or bias, in a model that determines how inputs are mapped to outputs.

Parameters are initialized before training and updated from data via optimization, such as gradient descent and backpropagation. They’re typically represented as tensors and may be fully trainable, frozen during fine-tuning, or modified via parameter-efficient adapters.

Unlike hyperparameters, which configure the model or training process but are not learned from data, parameters are learned.