Workflow profiling

Measure Workflow execution timing and find bottlenecks with Chrome-compatible profiling traces.

You can enable profiling to measure execution timing of your Workflow and identify performance bottlenecks. Profiling traces are compatible with Chrome's built-in tracing tool for visualization.

Example Workflow

The screenshots in this guide use the following workflow as an example; it includes 2x object detection, dynamic cropping, a Google Gemini call, and several visualization steps:

Example Workflow used for the profiling screenshots below

Profiling with the Inference SDK

To enable profiling when running workflows via the HTTP client, you first need to run self-hosted Inference Server with ENABLE_WORKFLOWS_PROFILING=True env variable. Then you can enable profiling when running workflows via the HTTP client by setting enable_profiling=True:

from inference_sdk import InferenceHTTPClient

client = InferenceHTTPClient(
    api_url="http://localhost:9001", # Serverless Cloud API does not support profiling
    api_key="API_KEY"
)

result = client.run_workflow(
    workspace_name="workspace-name",
    workflow_id="workflow-id",
    images={"image": "YOUR_IMAGE.jpg"},
    enable_profiling=True
)

When profiling is enabled, trace files are automatically saved to ./inference_profiling/ by default. Each trace file is named workflow_execution_tack_{timestamp}.json.

Viewing Profiling Traces

Profiling traces are saved in Chrome Tracing Format. To visualize them:

1. Open Chrome and navigate to chrome://tracing/.

2. Click Load and select your trace file (e.g., workflow_execution_tack_2026_03_30_10_46_04.json).

3. Use the timeline view to inspect execution timing for each workflow step.

Clicking on any span shows its duration, category, and arguments - such as which block's step_execution you are looking at:

Most of the time was spent on 2x object detection (process 57) and Gemini call (process 78). If you zoom-in, you can see other steps (detection offset, dynamic cropping, visualization, etc.) between the object detection steps:

Key Details

Detail	Value
Trace format	JSON, compatible with Chrome's chrome://tracing/ tool
Output location	./inference_profiling/ by default (configurable)
File naming	workflow_execution_tack_{YYYY_MM_DD_HH_MM_SS}.json

Limitations

• Compilation overhead: Workflow compilation adds 10-25ms latency, which is significant for small/fast models where GPU inference time is comparable. This overhead only occurs once at the start for video processing.

• Execution engine latency: The Workflows Execution Engine adds ~1-2ms nominal latency per frame.

• Dynamic blocks: Self-hosted deployments using dynamic blocks may experience +100-140ms latency per request due to pydantic model rebuilding. This is not applicable to the hosted platform or video processing where compilation happens once.

• Model block dominance: Typically ~95% of execution time is spent in model inference itself, which includes additional data transformations and metadata management.

Performance Tips

• Workflow definitions are cached for 15 minutes by default. Use use_cache=False in run_workflow() to bypass caching if needed.

• For video processing, compilation overhead is negligible since it only occurs once at the beginning.

• Profile your workflows to identify whether bottlenecks are in model inference, data transformations, or workflow orchestration.