Ultra‑Efficient Edge Inference

Optimize on-device inference for {{model}} on {{chipset}}. Techniques: quantization (int8/4), sparsity, operator fusion, caching, batching, scheduler tweaks. Report latency/energy tradeoffs and a rollout plan for older devices.