charts/hugs/values.yaml

image:
  registry: # Add the HUGS Registry that you subscribed to
  repository: # Add the HUGS Repository where the container is hosted in
  name: # Add the name of the HUGS container image
  pullPolicy: Always
  tag: latest

# serviceAccountName: hugs-service-account

podSecurityContext:
  {}
  # fsGroup: 2000

securityContext:
  {}
  # capabilities:
  #   drop:
  #   - ALL
  # readOnlyRootFilesystem: true
  # runAsNonRoot: true
  # runAsUser: 1000

env: {}
# Via `env` you can specify the port, defaults to 80.
# Note that if you update the exposed port, you should also
# update the service, readiness, and liveness ports too.
#   PORT: "80"

livenessProbe:
  enabled: true
  # You may want to increase the initialDelaySeconds for the bigger LLMs as Llama 3.1 405B
  # since the download will take longer and the default delay may not be enough for the
  # download to be completed
  # initialDelaySeconds: 360
  # periodSeconds: 15
  # timeoutSeconds: 5
  # failureThreshold: 3

readinessProbe:
  enabled: true
  # You may want to increase the initialDelaySeconds for the bigger LLMs as Llama 3.1 405B
  # since the download will take longer and the default delay may not be enough for the
  # download to be completed
  # initialDelaySeconds: 360
  # periodSeconds: 30
  # timeoutSeconds: 5
  # failureThreshold: 3

service:
  type: NodePort
  port: 80

# Optionally, one may want to start an ingress service too, with
# a custom IP to avoid having to forward the port to localhost
ingress:
  enabled: false
  # className: alb
  # annotations:
  #   alb.ingress.kubernetes.io/scheme: internet-facing
  # hosts:
  #   - host: ""
  #     paths:
  #       - path: /
  #         pathType: Prefix

resources:
  {}

volumes:
  - name: dshm
    emptyDir:
      medium: Memory
      sizeLimit: 1Gi

volumeMounts:
  - mountPath: /dev/shm
    name: dshm

nodeSelector:
  {}
  # eks.amazonaws.com/nodegroup: hugs-node-group

# https://kubernetes.io/docs/concepts/configuration/taint-and-toleration/
tolerations:
  {}

autoscaling:
  enabled: false
  # minReplicas: 1
  # maxReplicas: 2
  # targetMemoryUtilizationPercentage: ""
  # targetCPUUtilizationPercentage: ""