Skip to content
Spike Documentation

Inference workloads

Inference on Spike is deployed and managed for you by SCC AI Engineers. If a model must remain available behind a serving endpoint, discuss the requirement with your SCC AI Engineer. They will determine the appropriate deployment, resources, access controls, health checks, and cleanup policy with you.

What to prepare

Section titled “What to prepare”

Before meeting with the AI Engineer:

  1. Confirm that the model and its license are suitable for the project.
  2. Record the model revision, expected usage, and intended users or client applications.
  3. Prepare an immutable Harbor image tag, or discuss what the serving image needs to contain.
  4. Identify any runtime tokens or other secrets. These must be provided through the approved secret mechanism, never baked into an image or command line.
  5. Describe the expected request volume, response-time needs, and how long the endpoint is required.

The AI Engineer will create and validate the inference workload, including its compute-resource preset, replicas, endpoint, authentication, health checks, and request limits. Contact them again when the deployment needs to change or can be removed.