ZML is a production inference stack, purpose-built to decouple AI workloads from proprietary hardware.