Skip to main content

Simulation

Use Simulation to replay a trace dataset against a model service or an endpoint URL.

Create a simulation task​

  1. Open Simulation and click New Task.
  2. In Benchmark & Endpoint, choose Benchmark tool.
  3. Under Inference Endpoint, choose Model Service or Endpoint URL using the table below.
  4. Enter Model access token, then check the resolved Model.
  5. In Scenario & Prompt, select a scenario and Trace dataset. Click Download dataset if it is not downloaded.
  6. For a dataset with a timeline, select the replay start/end, scale factor and duration.
  7. In Load & Backend, review Stream responses. Add TTFT, TPOT or error-rate targets only if you want a pass/fail assessment.
  8. In Task Details, enter Task name and an optional description.
  9. Click Create Task. Open the new task card to follow its progress.
Endpoint choiceWhat to supply
Model ServiceA published, healthy service and a personal model access token
Endpoint URLThe endpoint address you want to test, plus a token if that service requires one

To create a service and token, follow Publish a model service. Leave Advanced backend settings unchanged for the first run.

With Endpoint URL, Lens validates the endpoint through its /v1/models API, using the optional model access token. You cannot continue or create a task while validation is pending or failed, even if you enter a model name manually. Correct the URL or token, or click Retry endpoint validation after restoring connectivity. The backend checks again before saving or scheduling a new task or rerun. DNS, connection, timeout, authentication, and invalid model-list responses block creation. A successful check does not guarantee that the endpoint will remain available during the run.

Addresses ending in .svc.cluster.local use Kubernetes cluster DNS and may not resolve from the Lens backend host. For a published cluster service, choose Model Service to use the cluster's benchmark execution path. Otherwise, supply an endpoint reachable from the Lens backend and benchmark runtime.

Read the result​

  1. Open the task card.
  2. In Summary, check the task state, completed requests and key metrics.
  3. Use Latency to inspect TTFT and TPOT over time, and Throughput to inspect processing rate.
  4. Open Errors & Logs to investigate failures.
  5. In Artifacts, click Download beside each file you want to keep.

Missing measurements are not zero. If a metric is unavailable, check Task Log for a failed run or collection problem before drawing conclusions.

Stop, rerun or delete a task​

GoalAction
Stop a running taskClick Stop on its card or detail page
Repeat a finished taskClick Rerun to create a new task with the same configuration
Remove a finished taskClick Delete, review the dialog, then click Delete Task

Download required artifacts before deletion.

Find an earlier task​

  1. Use search, status cards, Model or Dataset to narrow the list.
  2. For a date range or endpoint, open Advanced Filters, fill the fields and click Apply Filters.
  3. Use Rows per page and the pager for more results. Clear Filters resets the list.

For automated runs, see the Simulation API reference.