# Generated from run-config.json. Contains no credentials or cluster endpoints.
schema_version: 1
source: "run-config.json"
tested_configuration:
  topology:
    endpoint_picker_replicas: 1
    model_replicas: 1
    gpus_per_model_replica: 1
  endpoint_picker:
    image: "ghcr.io/llm-d/llm-d-router-endpoint-picker:v0.9.0"
    selected_detector: "request-count admission"
    max_concurrency: 128
    comparison_detectors:
      - "queue depth 2"
    selected_headroom: 0.15
  model_service:
    image: "registry.redhat.io/rhaii/vllm-cuda-rhel9@sha256:ad06abf3bb5235ebb5b2df84cd1b9fd09e823f0ff2eebfc82bb4590275ccfe0b"
    model: "openai/gpt-oss-20b"
    gpu: "1 NVIDIA H100"
    max_num_sequences: 128
    max_num_batched_tokens: 8192
    prefix_cache: "disabled"
  traffic:
    arrival_mode: "open-loop Poisson"
    pattern: "noisy sinusoidal phases with a timed surge and recovery"
    production_repeats: 3
    scenario: "batch isolation"
  metrics:
    captured_during_every_run: true
    direct_and_prometheus: true
  runner:
    scenario_source: "pipeline/benchmark.py"
    sha256: "3811ec26c46bf3a26fa643698ec54bf569bb4bc99c3ea22ca18f805cb077b8e0"
    traffic_driver: "GuideLLM"
    guidellm_version: "0.7.0"
    trace_compiler: "pipeline/guidellm_trace.py"
    launcher: "pipeline/run_guidellm_scenario.py"
    scenario_file: "scenario.json"