diff --git a/deployment/helm/charts/danswer/templates/inference-model-deployment.yaml b/deployment/helm/charts/danswer/templates/inference-model-deployment.yaml index 391a8e4289b..3897bba078e 100644 --- a/deployment/helm/charts/danswer/templates/inference-model-deployment.yaml +++ b/deployment/helm/charts/danswer/templates/inference-model-deployment.yaml @@ -25,6 +25,10 @@ spec: image: "{{ .Values.inferenceCapability.deployment.image.repository }}:{{ .Values.inferenceCapability.deployment.image.tag | default .Values.appVersionOverride | default .Chart.AppVersion }}" imagePullPolicy: {{ .Values.inferenceCapability.deployment.image.pullPolicy }} command: {{ toYaml .Values.inferenceCapability.deployment.command | nindent 14 }} + {{- if .Values.inferenceCapability.deployment.resources }} + resources: + {{- toYaml .Values.inferenceCapability.deployment.resources | nindent 10 }} + {{- end }} ports: - containerPort: {{ .Values.inferenceCapability.service.port }} envFrom: diff --git a/deployment/helm/charts/danswer/values.yaml b/deployment/helm/charts/danswer/values.yaml index de57f93ae6e..82e6ef892e5 100644 --- a/deployment/helm/charts/danswer/values.yaml +++ b/deployment/helm/charts/danswer/values.yaml @@ -39,6 +39,7 @@ inferenceCapability: tag: pullPolicy: IfNotPresent command: ["uvicorn", "model_server.main:app", "--host", "0.0.0.0", "--port", "9000"] + resources: port: 9000 volumeMounts: - name: inference-model-storage