Edge quantization brings 70B-class models to workstation deployments

Summary

A refreshed inference stack lowers memory requirements while keeping local retrieval, coding, and offline support workloads usable.