dynamo/docs/components/profiler/profiler-examples.md

title
Profiler Examples

Complete examples for profiling with DGDRs.

DGDR Examples

Dense Model: Rapid

Fast profiling (~30 seconds):

apiVersion: nvidia.com/v1beta1
kind: DynamoGraphDeploymentRequest
metadata:
  name: qwen-0-6b
spec:
  model: "Qwen/Qwen3-0.6B"
  image: "nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.0.0"

Dense Model: Thorough

Profiling with real GPU measurements:

apiVersion: nvidia.com/v1beta1
kind: DynamoGraphDeploymentRequest
metadata:
  name: vllm-dense-online
spec:
  model: "Qwen/Qwen3-0.6B"
  backend: vllm
  image: "nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.0.0"
  searchStrategy: thorough

MoE Model

Multi-node MoE profiling with SGLang:

[!IMPORTANT] The PVC referenced by modelCache.pvcName must already exist in the same namespace and contain the model weights at the specified pvcModelPath. The DGDR controller does not create or populate the PVC — it only mounts it into the profiling job and deployed workers.

apiVersion: nvidia.com/v1beta1
kind: DynamoGraphDeploymentRequest
metadata:
  name: sglang-moe
spec:
  model: "deepseek-ai/DeepSeek-R1"
  backend: sglang
  image: "nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.0.0"

  hardware:
    numGpusPerNode: 8

  modelCache:
    pvcName: "model-cache"
    pvcModelPath: "deepseek-r1"      # path within the PVC

Private Model

For gated or private HuggingFace models, pass your token via an environment variable injected into the profiling job. Create the secret first:

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

Then reference it in your DGDR:

apiVersion: nvidia.com/v1beta1
kind: DynamoGraphDeploymentRequest
metadata:
  name: llama-private
spec:
  model: "meta-llama/Llama-3.1-8B-Instruct"
  image: "nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.0.0"

  overrides:
    profilingJob:
      template:
        spec:
          containers: []    # required placeholder; leave empty to inherit defaults
          initContainers:
            - name: profiler
              env:
                - name: HF_TOKEN
                  valueFrom:
                    secretKeyRef:
                      name: hf-token-secret
                      key: HF_TOKEN

Custom SLA Targets

Control how the profiler optimizes your deployment by specifying latency targets and workload characteristics.

Explicit TTFT + ITL targets (default mode):

apiVersion: nvidia.com/v1beta1
kind: DynamoGraphDeploymentRequest
metadata:
  name: low-latency-dense
spec:
  model: "Qwen/Qwen3-0.6B"
  image: "nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.0.0"

  sla:
    ttft: 500      # Time To First Token target in milliseconds
    itl: 20        # Inter-Token Latency target in milliseconds

  workload:
    isl: 2000      # expected input sequence length (tokens)
    osl: 500       # expected output sequence length (tokens)

End-to-end latency target (alternative to ttft+itl):

spec:
  ...
  sla:
    e2eLatency: 10000    # total request latency budget in milliseconds

Optimization objective without explicit targets (maximize throughput or minimize latency):

spec:
  ...
  sla:
    optimizationType: throughput    # or: latency

Overrides

Use overrides to customize the profiling job pod spec — for example to add tolerations for GPU node taints or inject environment variables.

GPU node toleration (common on GKE and shared clusters):

apiVersion: nvidia.com/v1beta1
kind: DynamoGraphDeploymentRequest
metadata:
  name: dense-with-tolerations
spec:
  model: "Qwen/Qwen3-0.6B"
  image: "nvcr.io/nvidia/ai-dynamo/dynamo-frontend:1.0.0"

  overrides:
    profilingJob:
      template:
        spec:
          containers: []    # required placeholder; leave empty to inherit defaults
          tolerations:
            - key: nvidia.com/gpu
              operator: Exists
              effect: NoSchedule

Override the generated DynamoGraphDeployment (e.g., to use a custom worker image):

spec:
  ...
  overrides:
    dgd:
      apiVersion: nvidia.com/v1alpha1
      kind: DynamoGraphDeployment
      spec:
        services:
          VllmWorker:
            extraEnvs:
              - name: CUSTOM_ENV
                value: "my-value"

SGLang Runtime Profiling

Profile SGLang workers at runtime via HTTP endpoints:

# Start profiling
curl -X POST http://localhost:9090/engine/start_profile \
  -H "Content-Type: application/json" \
  -d '{"output_dir": "/tmp/profiler_output"}'

# Run inference requests to generate profiling data...

# Stop profiling
curl -X POST http://localhost:9090/engine/stop_profile

A test script is provided at examples/backends/sglang/test_sglang_profile.py:

python examples/backends/sglang/test_sglang_profile.py

View traces using Chrome's chrome://tracing, Perfetto UI, or TensorBoard.