@xiciyang/inference-serving

Deploy a model checkpoint as a reachable inference endpoint: probe environment, select backend, migrate if needed, deploy, health-check, register the endpoint, auto-degrade on failure, roll back, tear down — fully audited.