Optimizes trained neural networks for inference deployment with quantization and fusion techniques. Speeds up inference by 36x.