feat: add Kubernetes manifests for AI stack deployment including Qdrant, vLLM, and Triton

manifests/spacedock/ai-stack/ai-stack.yaml

@@ -136,7 +136,7 @@ spec:
             - "--served-model-name"
             - "Qwen2.5-7B-Instruct"
             - "--quantization"
-            - "awq_marlin"
+            - "awq"
             - "--dtype"
             - "half"
             - "--max-model-len"
@@ -205,6 +205,116 @@ spec:
       nodePort: 30800
 ---
 # ==============================================================================
+# 2B. vLLM GEMMA 4 12B INFERENCE ENGINE (snaisvr44hrvstr02 Tesla T4 GPU)
+# ==============================================================================
+apiVersion: v1
+kind: PersistentVolumeClaim
+metadata:
+  name: vllm-gemma4-models-pvc
+  namespace: ai-stack
+spec:
+  accessModes:
+    - ReadWriteOnce
+  storageClassName: longhorn
+  resources:
+    requests:
+      storage: 200Gi
+---
+apiVersion: apps/v1
+kind: Deployment
+metadata:
+  name: vllm-gemma4
+  namespace: ai-stack
+  labels:
+    app: vllm-gemma4
+spec:
+  replicas: 1
+  strategy:
+    type: Recreate
+  selector:
+    matchLabels:
+      app: vllm-gemma4
+  template:
+    metadata:
+      labels:
+        app: vllm-gemma4
+    spec:
+      nodeSelector:
+        kubernetes.io/hostname: snaisvr44hrvstr02
+      containers:
+        - name: vllm
+          image: vllm/vllm-openai:v0.32.0-cu129
+          imagePullPolicy: IfNotPresent
+          command: ["/bin/bash", "-c"]
+          args:
+            - |
+              rm -rf /usr/local/lib/python3.12/dist-packages/torchcodec*
+              exec vllm serve google/gemma-4-12B-it-qat-w4a16-ct \
+                --served-model-name Gemma-4-12B-it \
+                --dtype half \
+                --max-model-len 8192 \
+                --gpu-memory-utilization 0.92 \
+                --enforce-eager \
+                --enable-auto-tool-choice \
+                --tool-call-parser gemma4 \
+                --reasoning-parser gemma4 \
+                --host 0.0.0.0 \
+                --port 8000
+          env:
+            - name: HF_HOME
+              value: "/root/.cache/huggingface"
+            - name: VLLM_WORKER_MULTIPROC_METHOD
+              value: "spawn"
+          ports:
+            - name: http-openai
+              containerPort: 8000
+          resources:
+            requests:
+              cpu: "8"
+              memory: "24Gi"
+              nvidia.com/gpu: "1"
+            limits:
+              cpu: "24"
+              memory: "48Gi"
+              nvidia.com/gpu: "1"
+          readinessProbe:
+            httpGet:
+              path: /health
+              port: 8000
+            initialDelaySeconds: 30
+            periodSeconds: 10
+          volumeMounts:
+            - name: vllm-gemma4-models
+              mountPath: /root/.cache/huggingface
+            - name: dshm
+              mountPath: /dev/shm
+      volumes:
+        - name: vllm-gemma4-models
+          persistentVolumeClaim:
+            claimName: vllm-gemma4-models-pvc
+        - name: dshm
+          emptyDir:
+            medium: Memory
+            sizeLimit: 16Gi
+---
+apiVersion: v1
+kind: Service
+metadata:
+  name: vllm-gemma4
+  namespace: ai-stack
+  labels:
+    app: vllm-gemma4
+spec:
+  type: NodePort
+  selector:
+    app: vllm-gemma4
+  ports:
+    - name: http-openai
+      port: 8000
+      targetPort: 8000
+      nodePort: 30804
+---
+# ==============================================================================
 # 3. NVIDIA TRITON INFERENCE SERVER (Decision Engine & Multi-Model ML Serving)
 # ==============================================================================
 apiVersion: v1