The collection of all parameter values that define a trained AI model's behavior. Model weights are the "knowledge" encoded through training. When a model is deployed for inference, these weights are loaded into memory (often HBM) and used to process each input.
Discussed in Chapter 1 of This Is Server Country