changes to fix new image for vllm, embedding and reranker for cpu

ai-services/Makefile

@@ -1,6 +1,6 @@
 REGISTRY?=icr.io/ai-services-private
 IMAGE=ai-services
-TAG?=v0.0.368
+TAG?=v0.0.369
 CONTAINER_BUILDER?=podman
 CREDS_ARG := $(if $(and $(REGISTRY_USER),$(REGISTRY_PASSWORD)),--creds="$(REGISTRY_USER):$(REGISTRY_PASSWORD)")
 

ai-services/assets/applications/rag-cpu/podman/values.yaml

@@ -54,22 +54,22 @@ postgres:
 
 instruct:
   # @hidden
-  image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+  image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
   # @description API key for vLLM instruct service authentication. If empty, authentication is disabled. Provide a key to enable authentication.
   apiKey: ""
   model:
     # @hidden
-    name: ibm-granite/granite-3.3-8b-instruct
+    name: ibm-granite/granite-4.1-8b
     # @hidden
-    maxNumBatchedTokens: 26208
+    maxNumBatchedTokens: 8096
     # @hidden
-    maxModelLen: 26208
+    maxModelLen: 8096
     # @hidden
     maxBatchSize: 32
 
 embedding:
   # @hidden
-  image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+  image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
   model:
     # @hidden
     name: ibm-granite/granite-embedding-278m-multilingual
@@ -78,7 +78,7 @@ embedding:
 
 reranker:
   # @hidden
-  image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+  image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
   model:
     # @hidden
     name: BAAI/bge-reranker-v2-m3

ai-services/assets/applications/rag-dev/podman/values.yaml

@@ -67,7 +67,7 @@ instruct:
 
 embedding:
   # @hidden
-  image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+  image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
   model:
     # @hidden
     name: ibm-granite/granite-embedding-278m-multilingual
@@ -76,7 +76,7 @@ embedding:
 
 reranker:
   # @hidden
-  image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+  image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
   model:
     # @hidden
     name: BAAI/bge-reranker-v2-m3

ai-services/assets/applications/rag/podman/values.yaml

@@ -65,7 +65,7 @@ instruct:
 
 embedding:
   # @hidden
-  image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+  image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
   model:
     # @hidden
     name: ibm-granite/granite-embedding-278m-multilingual

ai-services/assets/catalog/openshift/values.yaml

@@ -9,7 +9,7 @@ ui:
       cpu: "500m"
 
 backend:
-  image: icr.io/ai-services-cicd/ai-services:v0.0.368
+  image: icr.io/ai-services-cicd/ai-services:v0.0.369
   runtime: "openshift"
   adminPasswordHash: ""
   # @generate:password length=32, special=false

ai-services/assets/catalog/podman/values.yaml

@@ -4,7 +4,7 @@ ui:
 
 backend:
   port: ""
-  image: icr.io/ai-services-cicd/ai-services:v0.0.368
+  image: icr.io/ai-services-cicd/ai-services:v0.0.369
   runtime: ""
   adminPasswordHash: ""
   # @generate:password length=32, special=false

ai-services/assets/components/embedding/vllm-cpu/podman/templates/embedding-server.yaml.tmpl

@@ -9,6 +9,10 @@ metadata:
 spec:
   restartPolicy: always
   volumes:
+    - name: dshm
+      emptyDir:
+        medium: Memory
+        sizeLimit: 512Mi
     - name: models
       hostPath:
         path: "{{ .BaseDir }}/models"
@@ -22,6 +26,7 @@ spec:
       - --served-model-name
       - {{ .Values.model }}
       - --port=8001
+      - --gpu-memory-utilization=0.5
       livenessProbe:
         httpGet:
           path: /health
@@ -35,12 +40,14 @@ spec:
           value: "16"
       resources:
         requests:
-          memory: "4Gi"
+          memory: "8Gi"
         limits:
-          memory: "4Gi"
+          memory: "8Gi"
       ports:
         - containerPort: 8001
       volumeMounts:
+        - mountPath: /dev/shm
+          name: dshm
         - mountPath: /models:z
           name: models
           readOnly: true

ai-services/assets/components/embedding/vllm-cpu/podman/values.yaml

@@ -1,3 +1,3 @@
-image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
 model: ""
 maxModelLen: 512

ai-services/assets/components/llm/vllm-cpu/openshift/values.schema.json

@@ -12,9 +12,14 @@
           "const": "ibm-granite/granite-3.3-8b-instruct",
           "title": "granite-3.3-8b-instruct",
           "description": "An 8b-parameter instruction-tuned LLM used for natural language understanding and generation tasks. In this project, it powers tasks such as prompt-based reasoning, summarization, answer generation, and other LLM-driven functionalities.\n\n**Model strengths**: Reliable instruction-following, natural conversation, strong reasoning, and enterprise-grade accuracy.\n\n**Supported Languages**: English (primary), Spanish, French, German, Portuguese, Italian, Japanese, Korean, Chinese, Arabic, and 100+ more."
+        },
+        {
+          "const": "ibm-granite/granite-4.1-8b",
+          "title": "granite-4.1-8b",
+          "description": "An 8b-parameter instruction-tuned LLM used for natural language understanding and generation tasks. In this project, it powers tasks such as prompt-based reasoning, summarization, answer generation, and other LLM-driven functionalities.\n\n**Model strengths**: Reliable instruction-following, natural conversation, strong reasoning, and enterprise-grade accuracy.\n\n**Supported Languages**: English (primary), Spanish, French, German, Portuguese, Italian, Japanese, Korean, Chinese, Arabic, and 100+ more."
         }
       ],
-      "default": "ibm-granite/granite-3.3-8b-instruct"
+      "default": "ibm-granite/granite-4.1-8b"
     },
     "apiKey": {
       "type": "string",

ai-services/assets/components/llm/vllm-cpu/podman/templates/vllm-server.yaml.tmpl

@@ -13,6 +13,10 @@ metadata:
 spec:
   restartPolicy: always
   volumes:
+    - name: dshm
+      emptyDir:
+        medium: Memory
+        sizeLimit: 512Mi
     - name: models
       hostPath:
         path: "{{ .BaseDir }}/models"
@@ -58,6 +62,8 @@ spec:
       ports:
         - containerPort: 8000
       volumeMounts:
+        - mountPath: /dev/shm
+          name: dshm
         - mountPath: /models:z
           name: models
           readOnly: true

ai-services/assets/components/llm/vllm-cpu/podman/values.schema.json

@@ -12,9 +12,14 @@
           "const": "ibm-granite/granite-3.3-8b-instruct",
           "title": "granite-3.3-8b-instruct",
           "description": "An 8b-parameter instruction-tuned LLM used for natural language understanding and generation tasks. In this project, it powers tasks such as prompt-based reasoning, summarization, answer generation, and other LLM-driven functionalities.\n\n**Model strengths**: Reliable instruction-following, natural conversation, strong reasoning, and enterprise-grade accuracy.\n\n**Supported Languages**: English (primary), Spanish, French, German, Portuguese, Italian, Japanese, Korean, Chinese, Arabic, and 100+ more."
+        },
+        {
+          "const": "ibm-granite/granite-4.1-8b",
+          "title": "granite-4.1-8b",
+          "description": "An 8b-parameter instruction-tuned LLM used for natural language understanding and generation tasks. In this project, it powers tasks such as prompt-based reasoning, summarization, answer generation, and other LLM-driven functionalities.\n\n**Model strengths**: Reliable instruction-following, natural conversation, strong reasoning, and enterprise-grade accuracy.\n\n**Supported Languages**: English (primary), Spanish, French, German, Portuguese, Italian, Japanese, Korean, Chinese, Arabic, and 100+ more."
         }
       ],
-      "default": "ibm-granite/granite-3.3-8b-instruct"
+      "default": "ibm-granite/granite-4.1-8b"
     },
     "apiKey": {
       "type": "string",

ai-services/assets/components/llm/vllm-cpu/podman/values.yaml

@@ -1,6 +1,6 @@
-image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
 model: ""
 apiKey: ""
-maxNumBatchedTokens: 26208
-maxModelLen: 26208
+maxNumBatchedTokens: 8096
+maxModelLen: 8096
 maxBatchSize: 32

ai-services/assets/components/reranker/vllm-cpu/podman/templates/reranker-server.yaml.tmpl

@@ -9,6 +9,10 @@ metadata:
 spec:
   restartPolicy: always
   volumes:
+    - name: dshm
+      emptyDir:
+        medium: Memory
+        sizeLimit: 512Mi
     - name: models
       hostPath:
         path: "{{ .BaseDir }}/models"
@@ -22,6 +26,8 @@ spec:
       - --served-model-name
       - {{ .Values.model }}
       - --port=8002
+      - --gpu-memory-utilization=0.5
+      - --enforce-eager
       livenessProbe:
         httpGet:
           path: /health
@@ -30,9 +36,6 @@ spec:
         periodSeconds: 30
         timeoutSeconds: 5
         failureThreshold: 3
-      env:
-        - name: OMP_NUM_THREADS
-          value: "8"
       resources:
         requests:
           memory: "5Gi"
@@ -41,6 +44,8 @@ spec:
       ports:
         - containerPort: 8002
       volumeMounts:
+        - mountPath: /dev/shm
+          name: dshm
         - mountPath: /models:z
           name: models
           readOnly: true

ai-services/assets/components/reranker/vllm-cpu/podman/values.yaml

@@ -1,2 +1,2 @@
-image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
 model: ""

ai-services/assets/prerequisites/openshift/serving-runtime-cpu/templates/servingruntime.yaml

@@ -25,7 +25,7 @@ spec:
     env:
     - name: HF_HOME
       value: /tmp/hf_home
-    image: icr.io/ppc64le-oss/vllm-ppc64le:0.19.1
+    image: icr.io/ppc64le-oss/vllm-ppc64le:0.28.0
     name: kserve-container
     ports:
     - containerPort: 8080

ai-services/assets/worker/openshift/values.yaml

@@ -1,5 +1,5 @@
 worker:
-  image: icr.io/ai-services-cicd/ai-services:v0.0.368
+  image: icr.io/ai-services-cicd/ai-services:v0.0.369
   runtime: "openshift"
   token: ""
   gatewayAddr: ""

ai-services/assets/worker/podman/values.yaml

@@ -1,5 +1,5 @@
 worker:
-  image: icr.io/ai-services-cicd/ai-services:v0.0.368
+  image: icr.io/ai-services-cicd/ai-services:v0.0.369
   runtime: "podman"
   token: ""
   gatewayAddr: ""