add-ml-service

To add a new KServe inference service to a LiftWing namespace, you append a new inference_services entry to the correct values.yaml and staging override files. This ensures consistent YAML structure without hand-editing fragile helmfile manifests.

Scaffolding a new namespace for a LiftWing ML service involves generating helmfile.yaml, values.yaml, and values-ml-staging-codfw.yaml from repo templates. It captures required container environment variables and configures network egress endpoints for the new environment.

Yes, you can configure request batching and replica settings for KServe deployments. The configuration process supports staging and production replica ranges alongside optional request batching parameters and CPU, memory, or GPU resource limits.

To ensure consistent YAML structure when editing helmfile values manifests, mirror the indentation and field ordering of existing entries. This approach applies when onboarding a new service into an existing namespace or creating a new one.

Yes, LiftWing namespace onboarding supports eqiad-only environments. When creating a brand-new namespace scaffold, the generation process can optionally include configuration and values support specifically for eqiad-only environments.