GPU route smoke (Restormel Testing)

Use a small suite to prove your BYO-GPU route is reachable, authorized, and returns a sane JSON shape after infra changes. Keep runners customer-funded (GitHub-hosted or self-hosted on your bill) to preserve margin on Testing as a product.

What to assert

  • HTTP — 200 (or your expected status) from the resolved endpoint through your app or gateway.
  • Latency — Wall clock under a generous bound (GPU cold start may be slow).
  • Schema — Response JSON includes expected keys (e.g. choices[0].message).
  • Auth failure — Wrong key yields 401/403, not 500 from your proxy.

Starter YAML fragment

Repo reference: docs/testing/gpu-route-smoke-template.yaml — copy into your restormel-testing.yaml and replace URLs, goals, and AC ids.

Acceptance criteria

Give each AC a stable id (e.g. gpu-smoke.reachable) so Release pack exports tie policy + route versions to human-readable proof. Run with testing run --ac gpu-smoke.reachable when you narrow scope.

Related