Recipe 03 — Quantize For Edge

1. Export a baseline artifact with export_onnx or export_tflite.
2. Run quantize_int8 with a representative calibration set.
3. Compare size and accuracy with quantize_report.
4. Benchmark the winning artifact with benchmark_latency.

Use this when the deployment question is not “can I quantize?” but “should I ship this quantized build?”