Bumping llm-d release to latest version v0.8.1

.github/workflows/ci-e2e-openshift.yaml

@@ -387,7 +387,8 @@ jobs:
       HPA_STABILIZATION_SECONDS: ${{ github.event.inputs.hpa_stabilization_seconds || '240' }}
       SKIP_CLEANUP: ${{ github.event.inputs.skip_cleanup || 'false' }}
       # Pin llm-d release for deterministic CI behavior
-      LLM_D_RELEASE: v0.7.0
+      LLM_D_RELEASE: v0.8.1
+      LLM_D_ROUTER_VERSION: v0.9.0
       GAIE_VERSION: v1.5.0
       # PR-specific namespaces for isolation between concurrent PR tests
       LLMD_NAMESPACE: llm-d-autoscaler-pr-${{ needs.gate.outputs.pr_number || github.run_id }}

.github/workflows/ci-pr-checks.yaml

@@ -326,9 +326,10 @@ jobs:
           USE_SIMULATOR: "true"
           SCALE_TO_ZERO_ENABLED: "false"
           CREATE_CLUSTER: "true"
-          # Emulated Kind: llm-d model serving is deployed by install-epp.sh (GAIE standalone chart + simulator).
+          # Emulated Kind: llm-d model serving is deployed by install-epp.sh (llm-d-router-standalone chart + simulator).
           # Pin llm-d release for deterministic CI behavior
-          LLM_D_RELEASE: "v0.7.0"
+          LLM_D_RELEASE: "v0.8.1"
+          LLM_D_ROUTER_VERSION: "v0.9.0"
           # Dual-controller smoke installs a secondary controller via Kustomize; overlay path must be explicit in CI.
           WVA_E2E_SECONDARY_OVERLAY_PATH: ${{ github.workspace }}/test/e2e/testdata/secondary-controller
           IMG: ${{ steps.build-image.outputs.image }}
@@ -424,7 +425,8 @@ jobs:
           USE_SIMULATOR: "true"
           SCALE_TO_ZERO_ENABLED: "true"
           CREATE_CLUSTER: "true"
-          LLM_D_RELEASE: "v0.7.0"
+          LLM_D_RELEASE: "v0.8.1"
+          LLM_D_ROUTER_VERSION: "v0.9.0"
           DEPLOY_LWS: "true"
           IMG: ${{ steps.build-image.outputs.image }}
           SKIP_BUILD: "true"

Makefile

@@ -24,7 +24,8 @@ ENVIRONMENT                 ?= kind-emulator
 USE_SIMULATOR               ?= true
 SCALE_TO_ZERO_ENABLED       ?= false
 SCALER_BACKEND              ?= prometheus-adapter  # prometheus-adapter (HPA), keda (ScaledObject), or none (skip, use pre-installed backend)
-LLM_D_RELEASE               ?= v0.7.0
+LLM_D_RELEASE               ?= v0.8.1
+LLM_D_ROUTER_VERSION        ?= v0.9.0
 GAIE_VERSION                ?= v1.5.0
 KV_SPARE_TRIGGER           ?=
 QUEUE_SPARE_TRIGGER         ?=
@@ -216,6 +217,7 @@ deploy-e2e-infra: ## Deploy e2e test infrastructure (WVA + EPP; no model server
 	fi
 	@ENVIRONMENT=$(ENVIRONMENT) \
 		LLM_D_RELEASE=$(LLM_D_RELEASE) \
+		LLM_D_ROUTER_VERSION=$(LLM_D_ROUTER_VERSION) \
 		GAIE_VERSION=$(GAIE_VERSION) \
 		LLMD_NS=$${LLMD_NS:-$(E2E_EMULATED_LLMD_NAMESPACE)} \
 		WVA_PROJECT=$(CURDIR) \

config/samples/hpa/co-ordinator/model-a-epp-values.yaml

@@ -1,58 +1,61 @@
-inferencePool:
+# Values overlay for the llm-d-router-standalone chart (oci://ghcr.io/llm-d/charts/llm-d-router-standalone).
+# Schema follows the chart's `router:` wrapper introduced with the project
+# rebrand (Inference Scheduler → llm-d Router) in llm-d v0.8.0 / chart v0.9.0.
+router:
+  inferencePool:
+    failureMode: FailOpen
   modelServers:
     matchLabels:
       llm-d.ai/model-id: model-a
-
-inferenceExtension:
-  image:
-    registry: ghcr.io
-    repository: llm-d/llm-d-inference-scheduler
-    tag: v0.8.0
-  replicas: 1
-  failureMode: FailOpen
-  flags:
-    v: 2
-  resources:
-    requests:
-      cpu: 100m
-      memory: 256Mi
-    limits:
-      cpu: 500m
-      memory: 512Mi
   extraServicePorts:
-  - name: http
-    port: 80
-    protocol: TCP
-    targetPort: 8081
-  monitoring:
-    prometheus:
-      enabled: true
-      auth:
-        enabled: false
-  pluginsConfigFile: model-a-plugins.yaml
-  pluginsCustomConfig:
-    model-a-plugins.yaml: |
-      apiVersion: inference.networking.x-k8s.io/v1alpha1
-      kind: EndpointPickerConfig
-      featureGates:
-      - flowControl
-      plugins:
-      - type: queue-scorer
-      - type: kv-cache-utilization-scorer
-      - type: prefix-cache-scorer
-      - type: no-hit-lru-scorer
-      schedulingProfiles:
-      - name: default
+    - name: http
+      port: 80
+      protocol: TCP
+      targetPort: 8081
+  epp:
+    image:
+      registry: ghcr.io
+      repository: llm-d/llm-d-router-endpoint-picker
+      tag: v0.9.0
+    replicas: 1
+    flags:
+      v: 2
+    resources:
+      requests:
+        cpu: 100m
+        memory: 256Mi
+      limits:
+        cpu: 500m
+        memory: 512Mi
+    monitoring:
+      prometheus:
+        enabled: true
+        auth:
+          enabled: false
+    pluginsConfigFile: model-a-plugins.yaml
+    pluginsCustomConfig:
+      model-a-plugins.yaml: |
+        apiVersion: llm-d.ai/v1alpha1
+        kind: EndpointPickerConfig
+        featureGates:
+        - flowControl
         plugins:
-        - pluginRef: queue-scorer
-          weight: 2
-        - pluginRef: kv-cache-utilization-scorer
-          weight: 2
-        - pluginRef: prefix-cache-scorer
-          weight: 3
-        - pluginRef: no-hit-lru-scorer
-          weight: 2
-  sidecar:
+        - type: queue-scorer
+        - type: kv-cache-utilization-scorer
+        - type: prefix-cache-scorer
+        - type: no-hit-lru-scorer
+        schedulingProfiles:
+        - name: default
+          plugins:
+          - pluginRef: queue-scorer
+            weight: 2
+          - pluginRef: kv-cache-utilization-scorer
+            weight: 2
+          - pluginRef: prefix-cache-scorer
+            weight: 3
+          - pluginRef: no-hit-lru-scorer
+            weight: 2
+  proxy:
     args:
     - --service-node
     - envoy-sidecar

config/samples/hpa/co-ordinator/model-b-epp-values.yaml

@@ -1,58 +1,61 @@
-inferencePool:
+# Values overlay for the llm-d-router-standalone chart (oci://ghcr.io/llm-d/charts/llm-d-router-standalone).
+# Schema follows the chart's `router:` wrapper introduced with the project
+# rebrand (Inference Scheduler → llm-d Router) in llm-d v0.8.0 / chart v0.9.0.
+router:
+  inferencePool:
+    failureMode: FailOpen
   modelServers:
     matchLabels:
       llm-d.ai/model-id: model-b
-
-inferenceExtension:
-  image:
-    registry: ghcr.io
-    repository: llm-d/llm-d-inference-scheduler
-    tag: v0.8.0
-  replicas: 1
-  failureMode: FailOpen
-  flags:
-    v: 2
-  resources:
-    requests:
-      cpu: 100m
-      memory: 256Mi
-    limits:
-      cpu: 500m
-      memory: 512Mi
   extraServicePorts:
-  - name: http
-    port: 80
-    protocol: TCP
-    targetPort: 8081
-  monitoring:
-    prometheus:
-      enabled: true
-      auth:
-        enabled: false
-  pluginsConfigFile: model-b-plugins.yaml
-  pluginsCustomConfig:
-    model-b-plugins.yaml: |
-      apiVersion: inference.networking.x-k8s.io/v1alpha1
-      kind: EndpointPickerConfig
-      featureGates:
-      - flowControl
-      plugins:
-      - type: queue-scorer
-      - type: kv-cache-utilization-scorer
-      - type: prefix-cache-scorer
-      - type: no-hit-lru-scorer
-      schedulingProfiles:
-      - name: default
+    - name: http
+      port: 80
+      protocol: TCP
+      targetPort: 8081
+  epp:
+    image:
+      registry: ghcr.io
+      repository: llm-d/llm-d-router-endpoint-picker
+      tag: v0.9.0
+    replicas: 1
+    flags:
+      v: 2
+    resources:
+      requests:
+        cpu: 100m
+        memory: 256Mi
+      limits:
+        cpu: 500m
+        memory: 512Mi
+    monitoring:
+      prometheus:
+        enabled: true
+        auth:
+          enabled: false
+    pluginsConfigFile: model-b-plugins.yaml
+    pluginsCustomConfig:
+      model-b-plugins.yaml: |
+        apiVersion: llm-d.ai/v1alpha1
+        kind: EndpointPickerConfig
+        featureGates:
+        - flowControl
         plugins:
-        - pluginRef: queue-scorer
-          weight: 2
-        - pluginRef: kv-cache-utilization-scorer
-          weight: 2
-        - pluginRef: prefix-cache-scorer
-          weight: 3
-        - pluginRef: no-hit-lru-scorer
-          weight: 2
-  sidecar:
+        - type: queue-scorer
+        - type: kv-cache-utilization-scorer
+        - type: prefix-cache-scorer
+        - type: no-hit-lru-scorer
+        schedulingProfiles:
+        - name: default
+          plugins:
+          - pluginRef: queue-scorer
+            weight: 2
+          - pluginRef: kv-cache-utilization-scorer
+            weight: 2
+          - pluginRef: prefix-cache-scorer
+            weight: 3
+          - pluginRef: no-hit-lru-scorer
+            weight: 2
+  proxy:
     args:
     - --service-node
     - envoy-sidecar

config/samples/hpa/co-ordinator/poc.mk

@@ -5,6 +5,8 @@ POC_DIR      := config/samples/hpa/co-ordinator
 POC_WVA_NS  := workload-variant-autoscaler-system
 POC_MON_NS  := workload-variant-autoscaler-monitoring
 GAIE_VERSION ?= v1.5.0
+LLM_D_RELEASE ?= v0.8.1
+LLM_D_ROUTER_VERSION ?= v0.9.0
 
 .PHONY: poc-install
 poc-install: ## [POC] Install per-model EPPs, sim workloads, gateway, and Prometheus Adapter rules
@@ -26,16 +28,16 @@ poc-install: ## [POC] Install per-model EPPs, sim workloads, gateway, and Promet
 	@echo ""
 	@echo "--- Step 2: Install model-a EPP (flowControl enabled) ---"
 	@helm upgrade --install model-a \
-	  oci://registry.k8s.io/gateway-api-inference-extension/charts/standalone \
-	  --version $(GAIE_VERSION) \
+	  oci://ghcr.io/llm-d/charts/llm-d-router-standalone \
+	  --version $(LLM_D_ROUTER_VERSION) \
 	  --namespace $(POC_NS) \
 	  -f $(POC_DIR)/model-a-epp-values.yaml
 	@echo ""
 	@echo "--- Step 3: Install model-b EPP (flowControl enabled) ---"
 	@kubectl delete configmap envoy -n $(POC_NS) --ignore-not-found=true 2>/dev/null; true
 	@helm upgrade --install model-b \
-	  oci://registry.k8s.io/gateway-api-inference-extension/charts/standalone \
-	  --version $(GAIE_VERSION) \
+	  oci://ghcr.io/llm-d/charts/llm-d-router-standalone \
+	  --version $(LLM_D_ROUTER_VERSION) \
 	  --namespace $(POC_NS) \
 	  -f $(POC_DIR)/model-b-epp-values.yaml
 	@echo ""

deploy/README.md

@@ -186,7 +186,7 @@ Options:
   -h, --help               Show help
 ```
 
-**llm-d stack** (gateway, EPP, ModelService): deploy using the [llm-d guides](https://github.com/llm-d/llm-d/tree/main/guides/optimized-baseline) directly. For EPP-only setup (GAIE standalone chart + tokenreview RBAC), use `deploy/install-epp.sh` after `install.sh`.
+**llm-d stack** (gateway, EPP, ModelService): deploy using the [llm-d guides](https://github.com/llm-d/llm-d/tree/main/guides/optimized-baseline) directly. For EPP-only setup (llm-d-router-standalone chart + tokenreview RBAC), use `deploy/install-epp.sh` after `install.sh`.
 
 **Environment variables** (see [Configuration Reference](#configuration-reference)):
 
@@ -214,8 +214,8 @@ export DEPLOY_OPERATIONAL_DASHBOARD=true    # Deploy Grafana and operational das
 
 ```bash
 ./deploy/install.sh -e kubernetes
-# EPP (GAIE standalone chart + RBAC):
-LLM_D_RELEASE=v0.7.0 GAIE_VERSION=v1.5.0 LLMD_NS=llm-d-optimized-baseline \
+# EPP (llm-d-router-standalone chart + RBAC):
+LLM_D_RELEASE=v0.8.1 LLM_D_ROUTER_VERSION=v0.9.0 GAIE_VERSION=v1.5.0 LLMD_NS=llm-d-optimized-baseline \
   ./deploy/install-epp.sh
 # Model server: follow llm-d/llm-d guides/optimized-baseline
 ```

deploy/ci-e2e-openshift/deploy-infrastructure.sh

@@ -35,13 +35,14 @@ kubectl patch deployment "$MODELSERVICE" -n "$LLMD_NAMESPACE" --type=json \
 [ -n "${VLLM_MAX_NUM_SEQS:-}" ] && kubectl patch deployment "$MODELSERVICE" -n "$LLMD_NAMESPACE" --type=json \
   -p="[{\"op\":\"add\",\"path\":\"/spec/template/spec/containers/0/args/-\",\"value\":\"--max-num-seqs=$VLLM_MAX_NUM_SEQS\"}]"
 
-# EPP / scheduler via install-epp.sh (handles GAIE standalone chart +
+# EPP / scheduler via install-epp.sh (handles llm-d-router-standalone chart +
 # flowControl feature gate + tokenreview RBAC).  llm-d is already
 # checked out at $GITHUB_WORKSPACE/llm-d so the script reuses it.
 WVA_PROJECT="$GITHUB_WORKSPACE" \
 LLMD_NS="$LLMD_NAMESPACE" \
 GAIE_VERSION="$GAIE_VERSION" \
 LLM_D_RELEASE="$LLM_D_RELEASE" \
+LLM_D_ROUTER_VERSION="$LLM_D_ROUTER_VERSION" \
 ENVIRONMENT=openshift \
 ENABLE_SCALE_TO_ZERO=true \
 SKIP_CLUSTER_CRDS=true \

deploy/install-epp.sh

@@ -1,18 +1,27 @@
 #!/usr/bin/env bash
 #
 # EPP infrastructure setup for all environments (kind-emulator, kubernetes, openshift).
-# Installs Gateway API CRDs, GAIE CRDs, and the GAIE standalone chart (EPP + InferencePool).
+# Installs Gateway API CRDs, GAIE CRDs, and the llm-d-router-standalone chart (EPP + InferencePool).
+# Three independent version axes:
+#   - LLM_D_RELEASE         pins the upstream llm-d/llm-d release whose guide
+#                           values files we fetch (router/base.values.yaml etc.)
+#   - LLM_D_ROUTER_VERSION  pins the llm-d-router chart + EPP image (released
+#                           from llm-d/llm-d-router on its own cadence)
+#   - GAIE_VERSION          pins the kubernetes-sigs Gateway API Inference
+#                           Extension CRDs (separate project)
 # For llm-d model serving, see the llm-d project guides at https://github.com/llm-d/llm-d.
 #
 # Usage:
-#   LLM_D_RELEASE=v0.8.0 GAIE_VERSION=v1.5.0 LLMD_NS=llm-d-sim ./deploy/install-epp.sh
+#   LLM_D_RELEASE=v0.8.1 LLM_D_ROUTER_VERSION=v0.9.0 GAIE_VERSION=v1.5.0 \
+#     LLMD_NS=llm-d-sim ./deploy/install-epp.sh
 #
 
 set -e
 set -o pipefail
 
 WVA_PROJECT=${WVA_PROJECT:-$PWD}
-LLM_D_RELEASE=${LLM_D_RELEASE:-v0.8.0}
+LLM_D_RELEASE=${LLM_D_RELEASE:-v0.8.1}
+LLM_D_ROUTER_VERSION=${LLM_D_ROUTER_VERSION:-v0.9.0}
 GAIE_VERSION=${GAIE_VERSION:-v1.5.0}
 LLMD_NS=${LLMD_NS:-llm-d-sim}
 ENVIRONMENT=${ENVIRONMENT:-kind-emulator}

deploy/kind-emulator/README.md

@@ -36,7 +36,7 @@ This deploys:
 
 - Kind cluster with 3 nodes, emulated GPUs (mixed vendors)
 - WVA controller
-- llm-d EPP (GAIE standalone)
+- llm-d EPP (llm-d-router-standalone chart)
 - Prometheus monitoring + Prometheus Adapter
 
 To also deploy a simulator model service for manual testing:

deploy/kind-emulator/install.sh

@@ -20,7 +20,7 @@ YELLOW='\033[1;33m'
 BLUE='\033[0;34m'
 NC='\033[0m' # No Color
 
-# Configuration (Kind emulator). EPP deploy is via deploy/install-epp.sh — set LLM_D_RELEASE / GAIE_VERSION / LLMD_NS there or in Makefile.
+# Configuration (Kind emulator). EPP deploy is via deploy/install-epp.sh — set LLM_D_RELEASE / LLM_D_ROUTER_VERSION / GAIE_VERSION / LLMD_NS there or in Makefile.
 WVA_PROJECT=${WVA_PROJECT:-$PWD}
 NAMESPACE_SUFFIX="sim"
 

deploy/kubernetes/README.md

@@ -62,7 +62,7 @@ This script automates the complete deployment process on kubernetes cluster incl
 export HF_TOKEN="your-hf-token-here"
 
 # Optional: Customize deployment (basename under llm-d guides/; llm-d main uses optimized-baseline)
-export GUIDE_NAME="optimized-baseline"                                  # Default for LLM_D_RELEASE v0.7.0+
+export GUIDE_NAME="optimized-baseline"                                  # Default for LLM_D_RELEASE v0.7.0+ (current default v0.8.1)
 export MODEL_ID="unsloth/Meta-Llama-3.1-8B"                               # Default
 export WVA_IMAGE_REPO="ghcr.io/llm-d/llm-d-workload-variant-autoscaler"         # Default
 export WVA_IMAGE_TAG="latest"                                             # Default