How do you optimize Keras models for low-latency inference?

Low-latency optimization reduces model size and computation cost.

Techniques include quantization, pruning, graph optimization, batch tuning, and hardware-specific compilation (TensorRT, TFLite).