Artificial intelligence model parameters are the adjustable numbers changed during training. In a neural network they include weights and biases that control how signals are combined across layers, allowing the model to encode patterns learned from data.
Parameter count is one indicator of model size and memory demand, but it is not a complete measure of capability. Architecture, training data, optimization, numerical precision and the quality of post-training can allow a smaller model to outperform a larger one on selected tasks.
Acronyms and aliases
AI model parameter synonymmodel parameter synonymmodel weight synonymartificial intelligence model parameter variantparameter variant
General terms
Specialised terms
Related terms
Frequently asked questions
What does a model's parameter count indicate?
It indicates how many learned numerical values the model contains, which affects storage and memory needs but does not by itself predict accuracy or usefulness.
Are model parameters changed during inference?
In ordinary inference the learned parameters remain fixed while inputs are processed. They change during training, fine-tuning or another explicit adaptation process.
