switch kokoro to remsky/Kokoro-FastAPI

apps/llama/kokoro.yaml

@@ -16,37 +16,22 @@ spec:
       labels:
         app: kokoro
     spec:
-      # Prevent Kubernetes from auto-injecting KOKORO_PORT=tcp://... (legacy
-      # Docker-link-style env vars from same-namespace Services), which Kokoro
-      # tries to parse as an integer and fails.
-      enableServiceLinks: false
       containers:
         - name: kokoro
-          # OpenAI-compatible Kokoro TTS server, CPU-only PyTorch backend
-          # Exposes POST /v1/audio/speech with multiple voices and streaming
-          image: hwdsl2/kokoro-server:latest@sha256:42886b8720e901f7e31aba2050cd03867767eb9f609bbc38fe93852e72f0feeb
+          # OpenAI-compatible Kokoro-FastAPI TTS server, CPU PyTorch backend.
+          # Models baked into the image (no PVC needed).
+          # v0.3.0 includes fix for per-request voice tensor memory leak (#459).
+          image: ghcr.io/remsky/kokoro-fastapi-cpu:v0.3.0
           ports:
             - containerPort: 8880
               name: http
               protocol: TCP
-          env:
-            # Default voice (can be overridden per-request)
-            - name: KOKORO_VOICE
-              value: "af_heart"
-          volumeMounts:
-            # Persistent cache for the ~320MB Kokoro model
-            - name: cache
-              mountPath: /var/lib/kokoro
           resources:
             requests:
               memory: "2Gi"
               cpu: "500m"
             limits:
               memory: "6Gi"
-      volumes:
-        - name: cache
-          persistentVolumeClaim:
-            claimName: kokoro-cache-lvmssd
 ---
 apiVersion: v1
 kind: Service