# Default values for gpustack. # This is a YAML-formatted file. # Declare variables to be passed into your templates. debug: false # registrationToken is used to authenticate workers with the server. # If null, a random 32-character token is generated on first install and reused on upgrades. registrationToken: null systemDefaultContainerRegistry: null image: repository: gpustack/gpustack tag: null pullPolicy: IfNotPresent server: ingress: # example: gpustack.example.com hostname: null tls: # if tls.cert and tls.key are set, the schema will be https cert: null # cert: |- # -----BEGIN CERTIFICATE----- # MIID... # -----END CERTIFICATE----- key: null # key: |- # -----BEGIN PRIVATE KEY----- # MIIE... # -----END PRIVATE KEY----- # if you have an existing database, set externalDatabaseURL to its connection string # e.g. postgresql://user:password@hostname:port/dbname externalDatabaseURL: null dataVolume: hostPath: null size: 10Gi apiPort: 30080 metricsPort: 10161 environmentConfig: {} clusterDomain: cluster.local gateway: # IngressClass name for the Higress gateway; GPUStack checks if it exists in the cluster to enable in-cluster gateway mode ingressClassname: higress # higress-core sub-chart configuration. # Set enabled: true to deploy Higress gateway alongside GPUStack. # If you already have Higress installed in the cluster, leave this disabled # and ensure gateway.ingressClassname matches your existing Higress instance. # Full list of available values: https://higress.io/helm-charts higress-core: enabled: true global: # Must match gateway.ingressClassname above ingressClass: higress # GPUStack manages its own plugin server; disable the built-in one enablePluginServer: false hub: docker.io/gpustack downstream: idleTimeout: 1800 upstream: idleTimeout: 3 gateway: hub: null image: mirrored-higress-gateway controller: hub: null image: mirrored-higress-higress pilot: hub: null image: mirrored-higress-pilot higressPlugins: image: repository: gpustack/higress-plugins pullPolicy: IfNotPresent tag: "0.2.0" worker: enabled: true # GPU vendor for the worker nodes. Controls driver volume mounts and runtimeClassName. # Supported values: # null - no vendor-specific driver mounts or runtimeClassName (e.g. CPU-only nodes) # nvidia - NVIDIA GPU; sets runtimeClassName to "nvidia" # mthreads - Moore Threads GPU; sets runtimeClassName to "mthreads" # amd - AMD GPU (ROCm); mounts /opt/rocm # ascend - Huawei Ascend NPU; mounts /usr/local/Ascend/driver and /usr/local/Ascend/ascend-toolkit # hygon - Hygon DCU; mounts /opt/hyhal and /opt/dtk # metax - MetaX GPU; mounts /opt/mxdriver and /opt/maca # iluvatar - Iluvatar GPU; mounts /usr/local/corex # cambricon - Cambricon MLU; mounts /usr/bin/cnmon and /usr/local/neuware # thead - T-Head PPU; mounts /usr/local/PPU_SDK gpuVendor: nvidia port: 10150 metricsPort: 10151 environmentConfig: {} # Host path for worker data directory. dataDir: /var/lib/gpustack extraVolumeMounts: [] # - name: custom-lib # mountPath: /opt/custom/lib # readOnly: true extraVolumes: [] # - name: custom-lib # hostPath: # path: /opt/custom/lib # type: DirectoryOrCreate