[MISC] Add routerconfig to runtimes and add pd runtimes

Makefile

@@ -2,10 +2,10 @@
 CHARTS_DIR := ./charts
 
 # Define the registry and image tagging
-REGISTRY     ?= ghcr.io/moirai-internal/ome
+REGISTRY     ?= ghcr.io/moirai-internal
 TAG          ?= $(GIT_TAG)
 ARCH         ?= linux/amd64
-MANAGER_IMG  ?= $(REGISTRY)/manager:$(TAG)
+MANAGER_IMG  ?= $(REGISTRY)/ome-manager:$(TAG)
 
 # Git version and commit information for build
 version_pkg = github.com/sgl-project/ome/pkg/version

charts/ome-resources/values.yaml

@@ -56,7 +56,7 @@ ome:
         cpu: 2
         memory: 4Gi
   omeAgent:
-    image: ghcr.io/moirai-internal/genai-ome-agent
+    image: ghcr.io/moirai-internal/ome-agent
     tag: *defaultVersion
     authType: InstancePrincipal
     compartmentId: ocid1.compartment.oc1..dummy-compartment

config/runtimes/kustomization.yaml

@@ -5,6 +5,9 @@ resources:
 - srt/deepseek-rdma-pd-rt.yaml
 - srt/deepseek-rdma-rt.yaml
 - srt/llama-4-maverick-17b-128e-instruct-fp8-rt.yaml
+- srt/llama-4-maverick-17b-128e-instruct-fp8-pd-rt.yaml
 - srt/llama-4-scout-17b-16e-instruct-rt.yaml
+- srt/llama-4-scout-17b-16e-instruct-pd-rt.yaml
 - srt/e5-mistral-7b-instruct-rt.yaml
 - srt/llama-3-3-70b-instruct-rt.yaml
+- srt/llama-3-3-70b-instruct-pd-rt.yaml

config/runtimes/srt/deepseek-rdma-pd-rt.yaml

@@ -1,7 +1,7 @@
 apiVersion: ome.io/v1beta1
 kind: ClusterServingRuntime
 metadata:
-  name: srt-deepseek-pd-rdma
+  name: srt-deepseek-rdma-pd
 spec:
   disabled: false
   modelSizeRange:
@@ -44,22 +44,9 @@ spec:
           effect: "NoSchedule"
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
-      nodeSelector:
-        oci.oraclecloud.com/rdma.authenticated: "16"
-        oci.oraclecloud.com/rdma.mlx_issues: "0"
-        oke.oraclecloud.com/pool.mode: cluster-network
-      affinity:
-        nodeAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-            nodeSelectorTerms:
-              - matchExpressions:
-                  - key: node.kubernetes.io/instance-type
-                    operator: In
-                    values:
-                      - BM.GPU.H100.8
       runner:
         name: ome-container
-        image: ghcr.io/moirai-internal/sgl:dev2
+        image: docker.io/lmsysorg/sglang:v0.4.8.post1-cu126
         env:
           - name: NVSHMEM_ENABLE_NIC_PE_MAPPING
             value: "1"
@@ -137,28 +124,15 @@ spec:
           timeoutSeconds: 30
     worker:
       size: 1
-      nodeSelector:
-        oci.oraclecloud.com/rdma.authenticated: "16"
-        oci.oraclecloud.com/rdma.mlx_issues: "0"
-        oke.oraclecloud.com/pool.mode: cluster-network
       tolerations:
         - key: nvidia.com/gpu
           operator: Exists
           effect: NoSchedule
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
-      affinity:
-        nodeAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-            nodeSelectorTerms:
-              - matchExpressions:
-                  - key: node.kubernetes.io/instance-type
-                    operator: In
-                    values:
-                      - BM.GPU.H100.8
       runner:
         name: ome-container
-        image: ghcr.io/moirai-internal/sgl:dev2
+        image: docker.io/lmsysorg/sglang:v0.4.8.post1-cu126
         command:
           - sh
           - -c
@@ -205,7 +179,7 @@ spec:
             value: "0"
           - name: NVSHMEM_ENABLE_NIC_PE_MAPPING
             value: "1"
-          - name: SGLANG_MOONCAKE_TRANS_THREAD
+          - name: SGLANG_DISAGGREGATION_THREAD_POOL_SIZE
             value: "8"
           - name: SGL_ENABLE_JIT_DEEPGEMM
             value: "1"
@@ -225,24 +199,11 @@ spec:
           effect: "NoSchedule"
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
-      nodeSelector:
-        oci.oraclecloud.com/rdma.authenticated: "16"
-        oci.oraclecloud.com/rdma.mlx_issues: "0"
-        oke.oraclecloud.com/pool.mode: cluster-network
-      affinity:
-        nodeAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-            nodeSelectorTerms:
-              - matchExpressions:
-                  - key: node.kubernetes.io/instance-type
-                    operator: In
-                    values:
-                      - BM.GPU.H100.8
       runner:
         name: ome-container
-        image: ghcr.io/moirai-internal/sgl:dev2
+        image: docker.io/lmsysorg/sglang:v0.4.8.post1-cu126
         env:
-          - name: SGLANG_MOONCAKE_TRANS_THREAD
+          - name: SGLANG_DISAGGREGATION_THREAD_POOL_SIZE
             value: "16"
           - name: SGL_ENABLE_JIT_DEEPGEMM
             value: "1"
@@ -309,28 +270,15 @@ spec:
           timeoutSeconds: 30
     worker:
       size: 1
-      nodeSelector:
-        oci.oraclecloud.com/rdma.authenticated: "16"
-        oci.oraclecloud.com/rdma.mlx_issues: "0"
-        oke.oraclecloud.com/pool.mode: cluster-network
       tolerations:
         - key: nvidia.com/gpu
           operator: Exists
           effect: NoSchedule
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
-      affinity:
-        nodeAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-            nodeSelectorTerms:
-              - matchExpressions:
-                  - key: node.kubernetes.io/instance-type
-                    operator: In
-                    values:
-                      - BM.GPU.H100.8
       runner:
         name: ome-container
-        image: ghcr.io/moirai-internal/sgl:dev2
+        image: docker.io/lmsysorg/sglang:v0.4.8.post1-cu126
         command:
           - sh
           - -c
@@ -372,7 +320,7 @@ spec:
             value: "0"
           - name: NVSHMEM_IB_TRAFFIC_CLASS
             value: "16"
-          - name: SGLANG_MOONCAKE_TRANS_THREAD
+          - name: SGLANG_DISAGGREGATION_THREAD_POOL_SIZE
             value: "16"
           - name: SGL_ENABLE_JIT_DEEPGEMM
             value: "1"

config/runtimes/srt/deepseek-rdma-rt.yaml

@@ -35,7 +35,7 @@ spec:
   routerConfig:
     runner:
       name: router
-      image: ghcr.io/moirai-internal/sgl-router:dev2
+      image: ghcr.io/moirai-internal/sgl-router:0.1.4.30f2a44
       resources:
         limits:
           cpu: "1"
@@ -75,30 +75,17 @@ spec:
           effect: "NoSchedule"
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
-      nodeSelector:
-        oci.oraclecloud.com/rdma.authenticated: "16"
-        oci.oraclecloud.com/rdma.mlx_issues: "0"
-        oke.oraclecloud.com/pool.mode: cluster-network
-      affinity:
-        nodeAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-            nodeSelectorTerms:
-              - matchExpressions:
-                  - key: node.kubernetes.io/instance-type
-                    operator: In
-                    values:
-                      - BM.GPU.H100.8
       runner:
         name: ome-container
-        image: ghcr.io/moirai-internal/sgl:dev2
+        image: docker.io/lmsysorg/sglang:v0.4.8.post1-cu126
         command:
           - sh
           - -c
           - >
             python3 -m sglang.launch_server 
             --host 0.0.0.0 --port 8080 
             --model-path ${MODEL_PATH} 
-            --tp 16 
+            --tp-size 16 
             --nccl-init $(LWS_LEADER_ADDRESS):5000 
             --nnodes ${LWS_GROUP_SIZE} 
             --node-rank ${LWS_WORKER_INDEX} 
@@ -139,28 +126,15 @@ spec:
           timeoutSeconds: 30
     worker:
       size: 1
-      nodeSelector:
-        oci.oraclecloud.com/rdma.authenticated: "16"
-        oci.oraclecloud.com/rdma.mlx_issues: "0"
-        oke.oraclecloud.com/pool.mode: cluster-network
       tolerations:
         - key: nvidia.com/gpu
           operator: Exists
           effect: NoSchedule
       dnsPolicy: ClusterFirstWithHostNet
       hostNetwork: true
-      affinity:
-        nodeAffinity:
-          requiredDuringSchedulingIgnoredDuringExecution:
-            nodeSelectorTerms:
-              - matchExpressions:
-                  - key: node.kubernetes.io/instance-type
-                    operator: In
-                    values:
-                      - BM.GPU.H100.8
       runner:
         name: ome-container
-        image: ghcr.io/moirai-internal/sgl:dev2
+        image: docker.io/lmsysorg/sglang:v0.4.8.post1-cu126
         command:
           - sh
           - -c
@@ -169,7 +143,7 @@ spec:
             --host 0.0.0.0 
             --port 8080 
             --model-path ${MODEL_PATH} 
-            --tp 16 
+            --tp-size 16 
             --nccl-init $(LWS_LEADER_ADDRESS):5000 
             --nnodes ${LWS_GROUP_SIZE} 
             --node-rank ${LWS_WORKER_INDEX} 

config/runtimes/srt/e5-mistral-7b-instruct-rt.yaml

@@ -22,7 +22,7 @@ spec:
   engineConfig:
     runner:
       name: ome-container
-      image: ghcr.io/moirai-internal/sgl:dev2
+      image: docker.io/lmsysorg/sglang:v0.4.8.post1-cu126
       ports:
         - containerPort: 8080
           name: http1
@@ -38,7 +38,7 @@ spec:
           --port=8080 \
           --enable-metrics \
           --model-path="$MODEL_PATH" \
-          --tp 1 \
+          --tp-size 1 \
           --is-embedding
       volumeMounts:
         - mountPath: /dev/shm
@@ -93,16 +93,4 @@ spec:
     volumes:
       - name: dshm
         emptyDir:
-          medium: Memory
-    affinity:
-      nodeAffinity:
-        requiredDuringSchedulingIgnoredDuringExecution:
-          nodeSelectorTerms:
-            - matchExpressions:
-                - key: node.kubernetes.io/instance-type
-                  operator: In
-                  values:
-                    - BM.GPU.B4.8
-                    - BM.GPU4.8
-                    - BM.GPU.A100-v2.8
-                    - BM.GPU.H100.8
+          medium: Memory