# Example ModelEndpoints
#
# Please review and edit where needed, e.g. any values starting with "my-"
#
# For detailed documentation of this API, use kubectl explain, e.g.
#   kubectl explain modelendpoint.spec
---
apiVersion: pais.vmware.com/v1alpha1
kind: ModelEndpoint
metadata:
  name: bge-small-en-v1.5-infinity
spec:
  model:
    ociRef: my-oci-registry.example.com/models/baai/bge-small-en:v1.5
  engine: Infinity
  type: Embeddings
  routingName: BAAI/bge-small-en-v1.5
  virtualMachineClassName: my-vm-class-choice-eg-best-effort-medium
  storageClassName: my-preferred-storage-class

---
apiVersion: pais.vmware.com/v1alpha1
kind: ModelEndpoint
metadata:
  name: mistral-7b-instruct-v0.2-vllm
spec:
  model:
    ociRef: my-oci-registry.example.com/more-models/mistralai/mistral-7b-instruct-v0.2@sha256:75eeddac5df9b8d8884c9ae32c5bda4e90d4d1c337829c654b97dad8d6eb72b2
    pullSecrets:
    # Optional, image pull secret if this OCI registry requires authentication
    - name: my-oci-registry-pull-secret
  engine: vLLM
  type: Completions
  routingName: mistralai/mistral-7b-instruct-v0.2
  virtualMachineClassName: my-vm-class-choice-eg-gpu-large
  storageClassName: my-preferred-storage-class