An optimization technique that speeds up how the model processes information, making it faster and more efficient without sacrificing quality.