Helm: values structure for multi-GPU-model serving?

[iacker]

I serve several models each on its own GPU profile. How should I structure Helm values to avoid copy-paste?

[iacker]

Model a list under .Values.models, each entry carrying name, image tag, gpuProfile, replicas and resource requests, then range over it in the templates to emit one Deployment+Service+ServiceMonitor per model. Keep cross-cutting defaults (registry, probes, DCGM labels) at the top level and merge with per-model overrides. This turns adding a model into a few lines of values instead of a new template, and keeps observability wiring consistent across all of them.