To bypass GPU memory requirement checks, annotate your workspace resource as follows:

apiVersion: kaito.sh/v1beta1
kind: Workspace
metadata:
  name: my-workspace
  annotations:
    kaito.sh/bypass-resource-checks: "True"
  # ... rest of the workspace spec

Caution: Using this annotation may lead to degraded performance or out-of-memory errors if GPU resources are insufficient for your model. Warnings will be logged accordingly.

For testing purposes, you may annotate your workspace resource with kaito.sh/enablelb: "True" to automatically create a LoadBalancer type service with a public IP for the inference service:

metadata:
  annotations:
    kaito.sh/enablelb: "True"

Important: This approach is NOT recommended for production environments. For production scenarios, use an Ingress Controller to safely expose the service.