Update the deployment to aigc

Signed-off-by: Le Yao <le.yao@intel.com>

Author: Le Yao

deployment/fastchat-api-server.yaml renamed to deployment/cse-aigc-api-server.yaml

@@ -14,7 +14,7 @@ spec:
     spec:
       containers:
       - name: fastchat-api-server
-        image: leyao/fastchat:latest
+        image: gar-registry.caas.intel.com/cpio/cnagc-fastchat-k8s:latest
         ports:
         - containerPort: 8000
         command: ["python3.9", "-m", "fastchat.serve.openai_api_server", "--controller-address", "http://fastchat-controller:21001", "--host", "0.0.0.0", "--port", "8000"]

deployment/fastchat-controller.yaml renamed to deployment/cse-aigc-controller.yaml

@@ -1,19 +1,19 @@
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: fastchat-controller
+  name: cse-aigc-controller
 spec:
   replicas: 1
   selector:
     matchLabels:
-      app: fastchat-controller
+      app: cse-aigc-controller
   template:
     metadata:
       labels:
-        app: fastchat-controller
+        app: cse-aigc-controller
     spec:
       containers:
-      - name: fastchat-controller
+      - name: cse-aigc-controller
         image: gar-registry.caas.intel.com/cpio/cnagc-fastchat-k8s:latest
         ports:
         - containerPort: 21001
@@ -27,10 +27,10 @@ spec:
 apiVersion: v1
 kind: Service
 metadata:
-  name: fastchat-controller
+  name: cse-aigc-controller
 spec:
   selector:
-    app: fastchat-controller
+    app: cse-aigc-controller
   ports:
     - protocol: TCP
       port: 21001

deployment/cse-aigc-monitor.yaml

@@ -0,0 +1,31 @@
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: cse-aigc-model-worker-non
+  namespace: non
+  labels:
+    service-monitor: cse-aigc-model-worker-non
+spec:
+  selector:
+    matchLabels:
+      app: cse-aigc-model-worker-non
+  endpoints:
+  - port: http
+    interval: 15s
+---
+
+apiVersion: monitoring.coreos.com/v1
+kind: ServiceMonitor
+metadata:
+  name: cse-aigc-model-worker-amx
+  namespace: amx
+  labels:
+    service-monitor: cse-aigc-model-worker-amx
+spec:
+  selector:
+    matchLabels:
+      app: cse-aigc-model-worker-amx
+  endpoints:
+  - port: http
+    interval: 15s
+

deployment/fastchat-web-ui.yaml renamed to deployment/cse-aigc-web-ui.yaml

@@ -1,19 +1,19 @@
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: fastchat-web-ui
+  name: cse-aigc-web-ui
 spec:
   replicas: 1
   selector:
     matchLabels:
-      app: fastchat-web-ui
+      app: cse-aigc-web-ui
   template:
     metadata:
       labels:
-        app: fastchat-web-ui
+        app: cse-aigc-web-ui
     spec:
       containers:
-      - name: fastchat-web-ui
+      - name: cse-aigc-web-ui
         image: gar-registry.caas.intel.com/cpio/cnagc-fastchat-k8s:latest
         ports:
         - containerPort: 9000
@@ -23,16 +23,18 @@ spec:
           value: "ui"
         - name: FASTCHAT_ROOT
           value: "/fastchat"
+        - name: CONTROLLER_SVC
+          value: cse-aigc-controller
 
 ---
 apiVersion: v1
 kind: Service
 metadata:
-  name: fastchat-web-ui
+  name: cse-aigc-web-ui
 spec:
   type: NodePort
   selector:
-    app: fastchat-web-ui
+    app: cse-aigc-web-ui
   ports:
     - protocol: TCP
       port: 9000

deployment/fastchat-model-worker-compare.yaml renamed to deployment/cse-aigc-worker-amx.yaml

@@ -1,19 +1,20 @@
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: fastchat-model-worker1
+  name: cse-aigc-model-worker-amx
+  namespace: amx
 spec:
   replicas: 1
   selector:
     matchLabels:
-      app: fastchat-model-worker1
+      app: cse-aigc-model-worker-amx
   template:
     metadata:
       labels:
-        app: fastchat-model-worker1
+        app: cse-aigc-model-worker-amx
     spec:
       containers:
-      - name: fastchat-model-worker1
+      - name: cse-aigc-model-worker-amx
         image: gar-registry.caas.intel.com/cpio/cnagc-fastchat-k8s:latest
         ports:
         - containerPort: 8000
@@ -24,29 +25,36 @@ spec:
           protocol: TCP
         env:
         - name: MODEL_NAME
-          value: "opt-1.3b-bf16-8b-samples"
+          value: "vicuna-7b-v1.3"
         - name: ATEN_CPU_CAPABILITY
-          value: "avx2"
+          value: "amx"
+        - name: CPU_ISA
+          value: "amx"
         - name: DEPLOY_TYPE
           value: "model"
         - name: FASTCHAT_ROOT
           value: "/fastchat"
         - name: MODEL_WORKER_SVC
-          value: fastchat-model-worker1
+          value: cse-aigc-model-worker-amx.amx.svc.cluster.local
+        - name: CONTROLLER_SVC
+          value: cse-aigc-controller.default.svc.cluster.local
         - name: MODEL_WORKER_PORT
           value: "21003"
+        - name: OMP_NUM_THREADS
+          value: "62"
         command: ["/entrypoint.sh"]
 
 ---
 apiVersion: v1
 kind: Service
 metadata:
-  name: fastchat-model-worker1
+  name: cse-aigc-model-worker-amx
+  namespace: amx
   labels:
-    app: fastchat-model-worker1
+    app: cse-aigc-model-worker-amx
 spec:
   selector:
-    app: fastchat-model-worker1
+    app: cse-aigc-model-worker-amx
   ports:
   - protocol: TCP
     name: model

deployment/fastchat-model-worker.yaml renamed to deployment/cse-aigc-worker-non.yaml

@@ -1,19 +1,20 @@
 apiVersion: apps/v1
 kind: Deployment
 metadata:
-  name: fastchat-model-worker
+  name: cse-aigc-model-worker-non
+  namespace: non
 spec:
   replicas: 1
   selector:
     matchLabels:
-      app: fastchat-model-worker
+      app: cse-aigc-model-worker-non
   template:
     metadata:
       labels:
-        app: fastchat-model-worker
+        app: cse-aigc-model-worker-non
     spec:
       containers:
-      - name: fastchat-model-worker
+      - name: cse-aigc-model-worker-non
         image: gar-registry.caas.intel.com/cpio/cnagc-fastchat-k8s:latest
         ports:
         - containerPort: 8000
@@ -25,28 +26,34 @@ spec:
         env:
         - name: MODEL_NAME
           value: "vicuna-7b-v1.3"
+          #value: "opt-1.3b-bf16-8b-samples"
         - name: ATEN_CPU_CAPABILITY
           value: "avx2"
         - name: DEPLOY_TYPE
           value: "model"
         - name: FASTCHAT_ROOT
           value: "/fastchat"
         - name: MODEL_WORKER_SVC
-          value: fastchat-model-worker
+          value: cse-aigc-model-worker-non.non.svc.cluster.local
         - name: MODEL_WORKER_PORT
           value: "21002"
+        - name: CONTROLLER_SVC
+          value: cse-aigc-controller.default.svc.cluster.local
+        - name: OMP_NUM_THREADS
+          value: "62"
         command: ["/entrypoint.sh"]
 
 ---
 apiVersion: v1
 kind: Service
 metadata:
-  name: fastchat-model-worker
+  name: cse-aigc-model-worker-non
+  namespace: non
   labels:
-    app: fastchat-model-worker
+    app: cse-aigc-model-worker-non
 spec:
   selector:
-    app: fastchat-model-worker
+    app: cse-aigc-model-worker-non
   ports:
     - protocol: TCP
       name: model