Train and deploy

Submit managed SFT, then verify actual success, checkpoints and resource release. Preserve artifacts through checkpoint resources, then follow the inference guide to register a compatible adapter and create authorized serving resources. Training success does not prove inference readiness; inference has independent permissions, budgets, charges and lifecycle.

Resource lifetime

Use Job stop/status for training, and each resource API for adapters/endpoints. Temporary scopes still manage supported resources; scope training factories are removed. The Console does not provide training submission or execution controls.