AI Glossary
Model parameters
model weights, parameters
Model parameters are the internal numerical values (weights) a model adjusts during training. They hold the learned knowledge, and their count is often given in billions.
- Weights are tuned during training on data.
- The number of parameters is usually given in billions (e.g. 7B, 70B).
- More parameters does not always mean a better model for a given task.
Parameters are the numbers inside a neural network, also called weights. During training the model gradually changes them to predict the next tokens more accurately on the training data. Once training is complete, it is the set of parameters that holds everything the model has learned.
The number of parameters is usually stated in billions, for example 7B or 70B. A larger model may have more capacity, but size does not determine quality — the data, the architecture, and the way it was tuned, for instance through fine-tuning, all matter too. A smaller, well-chosen model is often enough for a specific task.
Related terms
In guides