Skip to main content

Source: oceanid/docs/ML_BACKEND_CONNECTION.md | ✏️ Edit on GitHub

Label Studio ML Backend Connection

Connect the shared in‑cluster ML backend (ls-triton-adapter) to a project via the Label Studio UI or let the in‑cluster provisioner wire it automatically.

Steps​

  • Open your project
  • Go to: Settings → Model → Connect model
  • ML Backend URL: http://ls-triton-adapter.apps.svc.cluster.local:9090
  • Health: /health (200 OK)
  • Setup: /setup (GET/POST) — used by LS to validate the backend
  • Authentication: none
  • Optional: enable pre‑annotations for interactive predictions

Notes​

  • This is a per‑project configuration. There is no global auto‑connect.
  • The ls-triton-adapter service is shared and always running in the apps namespace.
  • The deprecated ls-ml-autoconnect service has been removed to avoid coupling infrastructure to user tokens.

Validation​

  • In project UI: Settings → Model → Test → should return 200 OK from /setup
  • From a pod: curl -s http://ls-triton-adapter.apps.svc.cluster.local:9090/health