feat: add TTS deployment to Helm chart

charts/smallest-self-host/Chart.yaml

@@ -16,7 +16,7 @@ type: application
 # to the chart and its templates, including the app version.
 # Versions are expected to follow Semantic Versioning (https://semver.org/)
 
-version: 0.2.2
+version: 0.3.0
 
 # This is the version number of the application being deployed. This version number should be
 # incremented each time you make changes to the application. Versions are not expected to

charts/smallest-self-host/templates/api-server-tts/config.yaml

@@ -0,0 +1,28 @@
+apiVersion: v1
+kind: ConfigMap
+metadata:
+  name: {{ .Values.apiServerTts.namePrefix }}-config
+  labels:
+    {{- include "smallest-self-host.labels" . | nindent 4 }}
+data:
+  api.toml: |
+    [license_proxy]
+      url = [
+        "http://{{ .Values.licenseProxy.namePrefix }}-internal:{{ .Values.licenseProxy.port}}{{ .Values.licenseProxy.baseUrl }}"
+      ]
+
+    [server]
+      base_url = "{{ .Values.apiServerTts.baseUrl }}"
+      host = "{{ .Values.apiServerTts.host }}"
+      port = {{ .Values.apiServerTts.port }}
+      lightning_v31_base_url = "http://{{ .Values.lightningTts.namePrefix }}-internal:{{ .Values.lightningTts.port}}{{ .Values.lightningTts.baseUrl }}"
+      {{- if .Values.redis.externalHost }}
+      redis_host = "{{ .Values.redis.externalHost }}"
+      {{- else }}
+      redis_host = "{{ .Release.Name }}-redis-master"
+      {{- end }}
+      redis_port = {{ .Values.redis.port }}
+      {{- if .Values.redis.auth.enabled }}
+      redis_password = "{{ .Values.redis.auth.password }}"
+      {{- end }}
+      redis_ssl = {{ .Values.redis.ssl }}

charts/smallest-self-host/templates/api-server-tts/deployment.yaml

@@ -0,0 +1,77 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: {{ .Values.apiServerTts.namePrefix }}
+  labels:
+    {{- include "smallest-self-host.labels" . | nindent 4 }}
+    app: {{ .Values.apiServerTts.namePrefix }}
+spec:
+  replicas: {{ .Values.scaling.replicas.apiServerTts }}
+  selector:
+    matchLabels:
+      app: {{ .Values.apiServerTts.namePrefix }}
+      {{- include "smallest-self-host.selectorLabels" . | nindent 6 }}
+  template:
+    metadata:
+      labels:
+        app: {{ .Values.apiServerTts.namePrefix }}
+        {{- include "smallest-self-host.selectorLabels" . | nindent 8 }}
+      {{- with .Values.apiServerTts.podAnnotations }}
+      annotations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+    spec:
+      {{- if or .Values.global.pullSecretRef .Values.global.imageCredentials.create }}
+      imagePullSecrets:
+      - name: {{ .Values.global.pullSecretRef | default (printf "%s-registry" (include "smallest-self-host.fullname" .)) }}
+      {{- end }}
+      containers:
+      - name: {{ .Values.apiServerTts.namePrefix }}
+        image: {{ .Values.apiServerTts.image }}
+        imagePullPolicy: {{ .Values.apiServerTts.imagePullPolicy }}
+        env:
+        - name: LICENSE_KEY
+          valueFrom:
+            secretKeyRef:
+              name: {{ .Values.global.licenseKeySecretRef | default (printf "%s-license" (include "smallest-self-host.fullname" .)) }}
+              key: license-key
+        volumeMounts:
+        - name: api-tts-config-volume
+          mountPath: /etc/config
+        ports:
+        - name: http
+          containerPort: {{ .Values.apiServerTts.port }}
+        livenessProbe:
+          httpGet:
+            path: /api/v1/health
+            port: {{ .Values.apiServerTts.port }}
+          initialDelaySeconds: {{ .Values.apiServerTts.livenessProbe.initialDelaySeconds }}
+          periodSeconds: {{ .Values.apiServerTts.livenessProbe.periodSeconds }}
+          timeoutSeconds: {{ .Values.apiServerTts.livenessProbe.timeoutSeconds }}
+        readinessProbe:
+          exec:
+            command:
+            - /bin/sh
+            - -c
+            - curl -sf http://localhost:{{ .Values.apiServerTts.port }}/healthz | grep -q "true"
+          initialDelaySeconds: {{ .Values.apiServerTts.readinessProbe.initialDelaySeconds }}
+          periodSeconds: {{ .Values.apiServerTts.readinessProbe.periodSeconds }}
+          timeoutSeconds: {{ .Values.apiServerTts.readinessProbe.timeoutSeconds }}
+          failureThreshold: {{ .Values.apiServerTts.readinessProbe.failureThreshold }}
+        startupProbe:
+          httpGet:
+            path: /api/v1/health
+            port: {{ .Values.apiServerTts.port }}
+          initialDelaySeconds: {{ .Values.apiServerTts.startupProbe.initialDelaySeconds }}
+          periodSeconds: {{ .Values.apiServerTts.startupProbe.periodSeconds }}
+          timeoutSeconds: {{ .Values.apiServerTts.startupProbe.timeoutSeconds }}
+          failureThreshold: {{ .Values.apiServerTts.startupProbe.failureThreshold }}
+        {{- if .Values.apiServerTts.resources }}
+        resources:
+          {{- toYaml .Values.apiServerTts.resources | nindent 10 }}
+        {{- end }}
+      volumes:
+      - name: api-tts-config-volume
+        configMap:
+          name: {{ .Values.apiServerTts.namePrefix }}-config
+      restartPolicy: Always

charts/smallest-self-host/templates/api-server-tts/service.yaml

@@ -0,0 +1,25 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ .Values.apiServerTts.namePrefix }}-external
+  labels:
+    {{- include "smallest-self-host.labels" . | nindent 4 }}
+    app: {{ .Values.apiServerTts.namePrefix }}
+spec:
+  type: {{ .Values.apiServerTts.service.type | default "ClusterIP" }}
+  selector:
+    app: {{ .Values.apiServerTts.namePrefix }}
+    {{- include "smallest-self-host.selectorLabels" . | nindent 4 }}
+  {{- if and (eq .Values.apiServerTts.service.type "LoadBalancer") .Values.apiServerTts.service.loadBalancerSourceRanges }}
+  loadBalancerSourceRanges:
+  {{- range .Values.apiServerTts.service.loadBalancerSourceRanges }}
+  - {{ . }}
+  {{- end }}
+  {{- end }}
+  {{- if and (eq .Values.apiServerTts.service.type "LoadBalancer") .Values.apiServerTts.service.externalTrafficPolicy }}
+  externalTrafficPolicy: {{ .Values.apiServerTts.service.externalTrafficPolicy }}
+  {{- end }}
+  ports:
+  - name: http
+    port: {{ .Values.apiServerTts.port }}
+    targetPort: {{ .Values.apiServerTts.port }}

charts/smallest-self-host/templates/lightning-asr/config.yaml

@@ -9,7 +9,7 @@ data:
     [server]
       port = {{ .Values.lightningAsr.port }}
     [model]
-      url = "{{ .Values.models.asrModelUrl }}"
+      url = "{{ .Values.lightningAsr.modelUrl | default .Values.models.asrModelUrl }}"
       path = "/smallest/models"
       model_name = "{{ .Values.lightningAsr.modelName }}"
 

charts/smallest-self-host/templates/lightning-tts/deployment.yaml

@@ -0,0 +1,88 @@
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: {{ .Values.lightningTts.namePrefix }}
+  labels:
+    {{- include "smallest-self-host.labels" . | nindent 4 }}
+    app: {{ .Values.lightningTts.namePrefix }}
+spec:
+  replicas: {{ .Values.scaling.replicas.lightningTts | default 1 }}
+  strategy:
+    type: RollingUpdate
+    rollingUpdate:
+      maxUnavailable: {{ .Values.lightningTts.updateStrategy.rollingUpdate.maxUnavailable }}
+      maxSurge: {{ .Values.lightningTts.updateStrategy.rollingUpdate.maxSurge }}
+  selector:
+    matchLabels:
+      app: {{ .Values.lightningTts.namePrefix }}
+      {{- include "smallest-self-host.selectorLabels" . | nindent 6 }}
+  template:
+    metadata:
+      labels:
+        app: {{ .Values.lightningTts.namePrefix }}
+        {{- include "smallest-self-host.selectorLabels" . | nindent 8 }}
+      {{- with .Values.lightningTts.podAnnotations }}
+      annotations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+    spec:
+      {{- if or .Values.global.pullSecretRef .Values.global.imageCredentials.create }}
+      imagePullSecrets:
+      - name: {{ .Values.global.pullSecretRef | default (printf "%s-registry" (include "smallest-self-host.fullname" .)) }}
+      {{- end }}
+      {{- with .Values.lightningTts.nodeSelector }}
+      nodeSelector:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      {{- with .Values.lightningTts.tolerations }}
+      tolerations:
+        {{- toYaml . | nindent 8 }}
+      {{- end }}
+      containers:
+      - name: {{ .Values.lightningTts.namePrefix }}
+        image: {{ .Values.lightningTts.image }}
+        imagePullPolicy: {{ .Values.lightningTts.imagePullPolicy | default "IfNotPresent" }}
+        env:
+        - name: LICENSE_KEY
+          valueFrom:
+            secretKeyRef:
+              name: {{ .Values.global.licenseKeySecretRef | default (printf "%s-license" (include "smallest-self-host.fullname" .)) }}
+              key: license-key
+        - name: PORT
+          value: "{{ .Values.lightningTts.port }}"
+        - name: REDIS_HOST
+          {{- if .Values.redis.externalHost }}
+          value: "{{ .Values.redis.externalHost }}"
+          {{- else }}
+          value: "{{ .Release.Name }}-redis-master"
+          {{- end }}
+        - name: REDIS_PORT
+          value: "{{ .Values.redis.port }}"
+        {{- if .Values.redis.auth.enabled }}
+        - name: REDIS_PASSWORD
+          value: "{{ .Values.redis.auth.password }}"
+        {{- end }}
+        ports:
+        - name: http
+          containerPort: {{ .Values.lightningTts.port }}
+          protocol: TCP
+        livenessProbe:
+          httpGet:
+            path: /health
+            port: {{ .Values.lightningTts.port }}
+          initialDelaySeconds: {{ .Values.lightningTts.livenessProbe.initialDelaySeconds }}
+          periodSeconds: {{ .Values.lightningTts.livenessProbe.periodSeconds }}
+          timeoutSeconds: {{ .Values.lightningTts.livenessProbe.timeoutSeconds }}
+        readinessProbe:
+          httpGet:
+            path: /health
+            port: {{ .Values.lightningTts.port }}
+          initialDelaySeconds: {{ .Values.lightningTts.readinessProbe.initialDelaySeconds }}
+          periodSeconds: {{ .Values.lightningTts.readinessProbe.periodSeconds }}
+          timeoutSeconds: {{ .Values.lightningTts.readinessProbe.timeoutSeconds }}
+          failureThreshold: {{ .Values.lightningTts.readinessProbe.failureThreshold }}
+        {{- if .Values.lightningTts.resources }}
+        resources:
+          {{- toYaml .Values.lightningTts.resources | nindent 10 }}
+        {{- end }}
+      restartPolicy: Always

charts/smallest-self-host/templates/lightning-tts/service.yaml

@@ -0,0 +1,17 @@
+apiVersion: v1
+kind: Service
+metadata:
+  name: {{ .Values.lightningTts.namePrefix }}-internal
+  labels:
+    {{- include "smallest-self-host.labels" . | nindent 4 }}
+    app: {{ .Values.lightningTts.namePrefix }}
+spec:
+  type: ClusterIP
+  selector:
+    app: {{ .Values.lightningTts.namePrefix }}
+    {{- include "smallest-self-host.selectorLabels" . | nindent 4 }}
+  ports:
+  - name: http
+    port: {{ .Values.lightningTts.port }}
+    targetPort: {{ .Values.lightningTts.port }}
+    protocol: TCP

charts/smallest-self-host/values.yaml

@@ -18,10 +18,8 @@ global:
 
 # use this if you want to share model volume across multiple pods, this prevents duplicate model downloads
 models:
-  # default model url for lightning-asr
-  # you can change this to the model url you want to use
-  # contact support@smallest.ai to get the model urls
-  asrModelUrl: "" # GET THIS FROM SUPPORT
+  # DEPRECATED: use lightningAsr.modelUrl instead. This field will be removed in a future version.
+  asrModelUrl: ""
   volumes:
     aws:
       efs:
@@ -33,6 +31,10 @@ scaling:
   replicas:
     # default replica count for lightning-asr
     lightningAsr: 1
+    # default replica count for lightning-tts
+    lightningTts: 1
+    # default replica count for api-server-tts
+    apiServerTts: 1
     # default replica count for license-proxy
     licenseProxy: 1
 
@@ -123,6 +125,9 @@ lightningAsr:
   baseUrl: ""
   podAnnotations: {}
 
+  # model url for lightning-asr, contact support@smallest.ai to get the model url
+  modelUrl: ""
+
   updateStrategy:
     rollingUpdate:
       maxUnavailable: 0
@@ -154,6 +159,88 @@ lightningAsr:
     requests:
       memory: 10Gi
 
+# ============ TTS STACK ============
+
+lightningTts:
+  namePrefix: lightning-tts
+  image: "quay.io/smallestinc/lightning-v3.1:latest"
+  imagePullPolicy: IfNotPresent
+  port: 8876
+  baseUrl: ""
+  podAnnotations: {}
+
+  # model url for lightning-tts, contact support@smallest.ai to get the model url
+  modelUrl: ""
+
+  updateStrategy:
+    rollingUpdate:
+      maxUnavailable: 0
+      maxSurge: 1
+
+  livenessProbe:
+    initialDelaySeconds: 300
+    periodSeconds: 30
+    timeoutSeconds: 10
+
+  readinessProbe:
+    initialDelaySeconds: 30
+    periodSeconds: 30
+    timeoutSeconds: 10
+    failureThreshold: 20
+
+  # we recommend using nodeSelector to schedule lightning-tts on GPU nodes
+  nodeSelector:
+
+  # we recommend using tolerations to allow lightning-tts to run on GPU nodes
+  tolerations:
+
+  labels: []
+
+  resources:
+    limits:
+      memory: 12Gi
+    requests:
+      memory: 10Gi
+
+apiServerTts:
+  namePrefix: api-server-tts
+  image: "quay.io/smallestinc/self-hosted-api-server:latest"
+  imagePullPolicy: IfNotPresent
+  port: 7100
+  baseUrl: ""
+  host: "0.0.0.0"
+  podAnnotations: {}
+
+  livenessProbe:
+    initialDelaySeconds: 30
+    periodSeconds: 10
+    timeoutSeconds: 5
+
+  readinessProbe:
+    initialDelaySeconds: 30
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 3
+
+  startupProbe:
+    initialDelaySeconds: 30
+    periodSeconds: 10
+    timeoutSeconds: 5
+    failureThreshold: 30
+
+  service:
+    type: ClusterIP
+    loadBalancerSourceRanges: []
+    externalTrafficPolicy: ""
+
+  resources:
+    requests:
+      cpu: 100m
+      memory: 256Mi
+    limits:
+      cpu: 500m
+      memory: 512Mi
+
 licenseProxy:
   namePrefix: license-proxy
   image: "quay.io/smallestinc/license-proxy:latest"