The initial phase of inference where the model processes the full input prompt before generating tokens.