[Chart] Enable Separate Resources for P and D

[robertgshaw2-redhat]

SUMMARY:

• previously, installer user same resources: for P and D
• this enables the help chart to have P and D specified separately

[robertgshaw2-redhat]

• Seems to be working with:

sampleApplication:
    baseConfigMapRefName: basic-gpu-with-nixl-preset
    model:
        modelArtifactURI: hf://RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic
        modelName: "RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic"
        auth:
            hfToken:
                name: llm-d-hf-token
                key: HF_TOKEN
    prefill:
        replicas: 4
        resources:
          limits:
            nvidia.com/gpu: 2
            rdma/ib: 1
          requests:
            nvidia.com/gpu: 2
            rdma/ib: 1
            cpu: "16"
            memory: 64Gi
        extraArgs:
            - "--tensor-parallel-size"
            - "2"
            - "--disable-log-requests"
            - "--max-model-len"
            - "32768"
            - "--block-size"
            - "128"
            - "--max-num-batched-tokens"
            - "32768"

    decode:
        replicas: 1
        resources:
          limits:
            nvidia.com/gpu: 8
            rdma/ib: 1
          requests:
            nvidia.com/gpu: 8
            rdma/ib: 1
            cpu: "16"
            memory: 64Gi
        extraArgs:
            - "--tensor-parallel-size"
            - "8"
            - "--disable-log-requests"
            - "--max-model-len"
            - "32768"
            - "--block-size"
            - "128"
modelservice:
  vllm:
    image:
      registry: docker.io    
      repository: robertgshaw2/vllm-nixl
      tag: het-tp-dev-0.1
  epp:
    defaultEnvVarsOverride:
      - name: ENABLE_KVCACHE_AWARE_SCORER
        value: "false"
      - name: ENABLE_PREFIX_AWARE_SCORER
        value: "true"
      - name: ENABLE_LOAD_AWARE_SCORER
        value: "true"
      - name: ENABLE_SESSION_AWARE_SCORER
        value: "false"
      - name: PD_ENABLED
        value: "true"
      - name: PD_PROMPT_LEN_THRESHOLD
        value: "10"
      - name: PREFILL_ENABLE_KVCACHE_AWARE_SCORER
        value: "false"
      - name: PREFILL_ENABLE_LOAD_AWARE_SCORER
        value: "true"
      - name: PREFILL_ENABLE_PREFIX_AWARE_SCORER
        value: "true"
      - name: PREFILL_ENABLE_SESSION_AWARE_SCORER
        value: "false"
redis:
  enabled: false

• installed with my justfile:

NAMESPACE := "pete-davidson"
MODEL := "RedHatAI/Llama-4-Scout-17B-16E-Instruct-FP8-dynamic"

logs POD:
    kubectl logs -f {{POD}} | grep -v "GET /metrics HTTP/1.1"

get-ips:
    just get-pods | awk '/^redhatai-llama-4-maverick-17b-128e-instruct-fp8-(decode|prefill)/ {print $6}'
get-pods:
    kubectl get pods -n {{NAMESPACE}} -o wide

hf-token:
  kubectl create secret generic hf-token-secret --from-literal=HF_TOKEN='$HF_TOKEN' -n {{NAMESPACE}}

[working-directory: '/home/rshaw/llm-d-deployer/quickstart']
install VALUES:
    ./llmd-installer.sh \
        --namespace {{NAMESPACE}} \
        --storage-class shared-vast --storage-size 300Gi \
        --values-file ./examples/rob-benchmarking/{{VALUES}}

start VALUES: 
    just install {{VALUES}} && \
    just hf-token && \
    just start-bench

[working-directory: '/home/rshaw/llm-d-deployer/quickstart']
uninstall VALUES:
    ./llmd-installer.sh \
        --namespace {{NAMESPACE}} \
        --storage-class shared-vast  --storage-size 300Gi \
        --values-file $PWD/../project/{{VALUES}} \
        --uninstall

gh-token GH_TOKEN:
    kubectl create secret generic gh-token-secret --from-literal=GH_TOKEN='{{GH_TOKEN}}' -n {{NAMESPACE}}

# Interactive benchmark commands:
start-bench:
    kubectl apply -n {{NAMESPACE}} -f benchmark-interactive-pod.yaml

delete-bench:
    kubectl delete pod -n {{NAMESPACE}} benchmark-interactive

exec-bench:
    kubectl cp reset_prefixes.sh {{NAMESPACE}}/benchmark-interactive:/app/reset_prefixes.sh && \
    kubectl cp Justfile.remote {{NAMESPACE}}/benchmark-interactive:/app/Justfile && \
    kubectl exec -it -n {{NAMESPACE}} benchmark-interactive -- /bin/bash