{
  "study": "vllm-batching-2026-10-15",
  "status": "complete",
  "conclusion": {
    "candidate": 4,
    "candidate_offered_rps": 0.625,
    "confirmation_stats": {
      "1": {
        "completed": 3000,
        "every_run_passes": true,
        "mean_output_tokens_per_second": 71.18377930461065,
        "std_output_tokens_per_second": 0.0007531740083949376,
        "min_output_tokens_per_second": 71.18328068281232,
        "max_output_tokens_per_second": 71.18464570813948,
        "p99_seconds": [
          1.8757182321192523,
          1.9332565942063227,
          1.891327634578629
        ]
      },
      "4": {
        "completed": 3000,
        "every_run_passes": true,
        "mean_output_tokens_per_second": 79.98575305600694,
        "std_output_tokens_per_second": 0.001148370153672993,
        "min_output_tokens_per_second": 79.98473569099477,
        "max_output_tokens_per_second": 79.98699827115578,
        "p99_seconds": [
          2.102773748633626,
          2.126980768921494,
          2.14368828322622
        ]
      }
    },
    "selected_confirmation_runs": {
      "1": [
        "confirm-s1-r1",
        "confirm-s1-r2-lower-rate",
        "confirm-s1-r3-lower-rate"
      ],
      "4": [
        "confirm-s4-r1-lower-rate",
        "confirm-s4-r2-lower-rate",
        "confirm-s4-r3-lower-rate"
      ]
    },
    "mean_improvement_fraction": 0.12365139695281924,
    "improvement_demonstrated": true,
    "frozen_limit_seconds": 2.260152011719183,
    "best_passing_s1_output_tokens_per_second": 71.19296360176575,
    "rule": "Every selected operating-point confirmation passes; all failed higher-rate attempts retained; candidate minimum >=1.1 times best passing seqs=1 and exceeds observed throughput ranges",
    "complete": true
  },
  "calibration": {
    "experimental_threshold_not_customer_slo": true,
    "baseline_run": "threshold-baseline-s1",
    "baseline_completed": 1000,
    "baseline_p99_s": 1.883460009765986,
    "multiplier": 1.2,
    "limit_s": 2.260152011719183,
    "frozen_at_epoch": 1792078116.3944955,
    "light_rate": 0.28058985357823446
  },
  "configuration": {
    "model": "nvidia/Qwen3.6-35B-A3B-NVFP4",
    "model_revision": "1355db6a052410cfd62085d94b58866fd0f2c3c5",
    "image": "nvcr.io/nvidia/ai-dynamo/vllm-runtime:1.5.1@sha256:2b224a823d52a198ec0eeb0e3e4f5cf26cd8a7fc2e9812ee16204b6b1e974870",
    "vllm_version": "0.28.0",
    "weight_precision": "NVFP4, Marlin weight compression",
    "compute_dtype": "bfloat16",
    "kv_cache_dtype": "fp8",
    "max_num_seqs": [
      1,
      2,
      4,
      8
    ],
    "max_num_batched_tokens": 2048,
    "max_model_len": 4096,
    "prefix_caching": false,
    "speculative_decoding": false,
    "kv_cache_memory_bytes": 2147483648,
    "seed": 20261015,
    "workload_sha256": "04d1926cd4cead96317d1f623b7a6f5b9277bc680a02e1ac4476dde836e95a92",
    "dataset_count": 30000,
    "arrivals": "deterministic open-loop; no client concurrency cap",
    "hardware": {
      "accelerator": "NVIDIA GB10",
      "gpus_per_worker": 1,
      "cpu_cores": "20",
      "node_memory_capacity": "125446488Ki",
      "architecture": "arm64",
      "driver_version": "580.95.05",
      "memory_reporting": "GB10 unified memory; dedicated nvidia-smi memory.total/free unavailable, host MemAvailable recorded instead"
    }
  },
  "runs": [
    {
      "name": "confirm-s1-r1",
      "seqs": 1,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5561203213591365,
      "actual_dispatch_rps": 0.5561203185996528,
      "achieved_rps": 0.5561300445948397,
      "output_tokens_per_second": 71.18464570813948,
      "observed_output_tokens_per_second_all_statuses": 71.18464570813948,
      "e2e_p50_s": 1.7735471457999665,
      "e2e_p95_s": 1.8138533901881602,
      "e2e_p99_s": 1.8757182321192523,
      "ttft_p99_s": 0.2135663418614422,
      "itl_p99_s": 0.0244678258964268,
      "per_request_mean_itl_p99_s": 0.013433520518422976,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002219700081332121,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 93.0548128342246,
      "gpu_temp_max_c": 74.0,
      "gpu_power_mean_w": 37.432580882352944,
      "host_mem_available_min_bytes": 84020449280.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "confirm-s1-r2-lower-rate",
      "seqs": 1,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5561203213591365,
      "actual_dispatch_rps": 0.5561202815786025,
      "achieved_rps": 0.5561193803344713,
      "output_tokens_per_second": 71.18328068281232,
      "observed_output_tokens_per_second_all_statuses": 71.18328068281232,
      "e2e_p50_s": 1.7842247294975095,
      "e2e_p95_s": 1.83968079824117,
      "e2e_p99_s": 1.9332565942063227,
      "ttft_p99_s": 0.306840058703674,
      "itl_p99_s": 0.024518461343686782,
      "per_request_mean_itl_p99_s": 0.013365692415813903,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0021620250830892474,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 1.0,
      "max_waiting": 1.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 92.27837116154873,
      "gpu_temp_max_c": 72.0,
      "gpu_power_mean_w": 37.085859145527365,
      "host_mem_available_min_bytes": 84047044608.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "confirm-s1-r3-lower-rate",
      "seqs": 1,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5561203213591365,
      "actual_dispatch_rps": 0.5561216049740211,
      "achieved_rps": 0.5561204025225011,
      "output_tokens_per_second": 71.18341152288014,
      "observed_output_tokens_per_second_all_statuses": 71.18341152288014,
      "e2e_p50_s": 1.78524627238221,
      "e2e_p95_s": 1.8312933041372161,
      "e2e_p99_s": 1.891327634578629,
      "ttft_p99_s": 0.22866745461476962,
      "itl_p99_s": 0.024221226148074385,
      "per_request_mean_itl_p99_s": 0.01348109731652914,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0021571057150140405,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 1.0,
      "max_waiting": 1.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 93.46461949265688,
      "gpu_temp_max_c": 72.0,
      "gpu_power_mean_w": 37.62771161548732,
      "host_mem_available_min_bytes": 84093702144.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "confirm-s4-r1",
      "seqs": 4,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6903562609975488,
      "actual_dispatch_rps": 0.6903591237719808,
      "achieved_rps": 0.6901169863355783,
      "output_tokens_per_second": 88.33497425095402,
      "observed_output_tokens_per_second_all_statuses": 88.33497425095402,
      "e2e_p50_s": 2.163195592678676,
      "e2e_p95_s": 2.2100665919031597,
      "e2e_p99_s": 2.2883242940405033,
      "ttft_p99_s": 0.29924553370641654,
      "itl_p99_s": 0.061866013686812944,
      "per_request_mean_itl_p99_s": 0.016419308217993787,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002156270637642592,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.02244389027432,
      "gpu_temp_max_c": 75.0,
      "gpu_power_mean_w": 39.31270157938487,
      "host_mem_available_min_bytes": 84007264256.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "confirm-s4-r1-lower-rate",
      "seqs": 4,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.625,
      "actual_dispatch_rps": 0.6249999282145507,
      "achieved_rps": 0.6248984239934046,
      "output_tokens_per_second": 79.98699827115578,
      "observed_output_tokens_per_second_all_statuses": 79.98699827115578,
      "e2e_p50_s": 2.017388108506566,
      "e2e_p95_s": 2.0503460819090833,
      "e2e_p99_s": 2.102773748633626,
      "ttft_p99_s": 0.23338506419386246,
      "itl_p99_s": 0.06055761029172562,
      "per_request_mean_itl_p99_s": 0.015083734917959269,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0021991423916188067,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 93.86716981132075,
      "gpu_temp_max_c": 72.0,
      "gpu_power_mean_w": 38.04933735849056,
      "host_mem_available_min_bytes": 83791867904.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "confirm-s4-r2-lower-rate",
      "seqs": 4,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.625,
      "actual_dispatch_rps": 0.6250002957630908,
      "achieved_rps": 0.6248807475858966,
      "output_tokens_per_second": 79.98473569099477,
      "observed_output_tokens_per_second_all_statuses": 79.98473569099477,
      "e2e_p50_s": 2.020176332014671,
      "e2e_p95_s": 2.05571279371361,
      "e2e_p99_s": 2.126980768921494,
      "ttft_p99_s": 0.24669390359515087,
      "itl_p99_s": 0.06047702644078525,
      "per_request_mean_itl_p99_s": 0.015178175435956533,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0024902189285785424,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.54833836858006,
      "gpu_temp_max_c": 76.0,
      "gpu_power_mean_w": 39.295883685800604,
      "host_mem_available_min_bytes": 83958538240.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "confirm-s4-r3-lower-rate",
      "seqs": 4,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.625,
      "actual_dispatch_rps": 0.624999956994892,
      "achieved_rps": 0.6248869156708613,
      "output_tokens_per_second": 79.98552520587025,
      "observed_output_tokens_per_second_all_statuses": 79.98552520587025,
      "e2e_p50_s": 2.0525810764956987,
      "e2e_p95_s": 2.0859264768987487,
      "e2e_p99_s": 2.14368828322622,
      "ttft_p99_s": 0.2546573100067325,
      "itl_p99_s": 0.06014324169387686,
      "per_request_mean_itl_p99_s": 0.015354923257794676,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002191123167576734,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.68117029257314,
      "gpu_temp_max_c": 73.0,
      "gpu_power_mean_w": 39.116540885221305,
      "host_mem_available_min_bytes": 83815960576.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "pilot-s1",
      "seqs": 1,
      "requested": 20,
      "launched": 20,
      "completed": 20,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.49999239133959456,
      "achieved_rps": 0.5029317208318012,
      "output_tokens_per_second": 64.37526026647055,
      "observed_output_tokens_per_second_all_statuses": 64.37526026647055,
      "e2e_p50_s": 1.7819603724929038,
      "e2e_p95_s": 1.7937063335550192,
      "e2e_p99_s": 1.7956720363114436,
      "ttft_p99_s": 0.1387014873116277,
      "itl_p99_s": 0.02222867218646571,
      "per_request_mean_itl_p99_s": 0.0130738236582068,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0020339742739452047,
      "output_length_counts": {
        "128": 20
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 9.235294117647058,
      "gpu_temp_max_c": 57.0,
      "gpu_power_mean_w": 25.370823529411762,
      "host_mem_available_min_bytes": 84227530752.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "refine-s1-0",
      "seqs": 1,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6008656345719405,
      "actual_dispatch_rps": 0.6008642526243859,
      "achieved_rps": 0.5657558648446345,
      "output_tokens_per_second": 72.41675070011321,
      "observed_output_tokens_per_second_all_statuses": 72.41675070011321,
      "e2e_p50_s": 6.876371457627101,
      "e2e_p95_s": 11.504488392052007,
      "e2e_p99_s": 11.89387766868822,
      "ttft_p99_s": 10.2349497047333,
      "itl_p99_s": 0.025859654497762695,
      "per_request_mean_itl_p99_s": 0.01316178038539287,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002787325955141571,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 1.0,
      "max_waiting": 6.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 94.93877551020408,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 37.14081632653061,
      "host_mem_available_min_bytes": 83927306240.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "refine-s1-1",
      "seqs": 1,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5561203213591365,
      "actual_dispatch_rps": 0.5561197718070949,
      "achieved_rps": 0.556195028138795,
      "output_tokens_per_second": 71.19296360176575,
      "observed_output_tokens_per_second_all_statuses": 71.19296360176575,
      "e2e_p50_s": 1.772036203132302,
      "e2e_p95_s": 1.810287220527971,
      "e2e_p99_s": 1.8511582712347445,
      "ttft_p99_s": 0.18659801347166705,
      "itl_p99_s": 0.02476287353973026,
      "per_request_mean_itl_p99_s": 0.013374521319574988,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.001909901180188172,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.08,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 37.894679999999994,
      "host_mem_available_min_bytes": 83937218560.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "refine-s1-2",
      "seqs": 1,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5784929779655386,
      "actual_dispatch_rps": 0.5784940312859098,
      "achieved_rps": 0.5663672248721092,
      "output_tokens_per_second": 72.49500478362998,
      "observed_output_tokens_per_second_all_statuses": 72.49500478362998,
      "e2e_p50_s": 3.5949577735664207,
      "e2e_p95_s": 5.248396190059429,
      "e2e_p99_s": 5.399896986836539,
      "ttft_p99_s": 3.7366481254258543,
      "itl_p99_s": 0.030511738720961176,
      "per_request_mean_itl_p99_s": 0.013890603574618014,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.001975484387075994,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 1.0,
      "max_waiting": 2.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 36.642639455782316,
      "host_mem_available_min_bytes": 84125011968.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "refine-s2-0",
      "seqs": 2,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6008656345719405,
      "actual_dispatch_rps": 0.60087007932121,
      "achieved_rps": 0.6002582602100545,
      "output_tokens_per_second": 76.83305730688697,
      "observed_output_tokens_per_second_all_statuses": 76.83305730688697,
      "e2e_p50_s": 1.9498673899215646,
      "e2e_p95_s": 1.9766540484146389,
      "e2e_p99_s": 1.9929395572180513,
      "ttft_p99_s": 0.16312767919982468,
      "itl_p99_s": 0.050259020606608834,
      "per_request_mean_itl_p99_s": 0.014491995820377112,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0016963057148677793,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.82142857142857,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 39.36861428571429,
      "host_mem_available_min_bytes": 84078182400.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "refine-s2-1",
      "seqs": 2,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6456109477847447,
      "actual_dispatch_rps": 0.6456239940861239,
      "achieved_rps": 0.6442966086877225,
      "output_tokens_per_second": 82.46996591202848,
      "observed_output_tokens_per_second_all_statuses": 82.46996591202848,
      "e2e_p50_s": 2.0509604113467503,
      "e2e_p95_s": 2.08853172734016,
      "e2e_p99_s": 2.128729665953433,
      "ttft_p99_s": 0.25330576753738593,
      "itl_p99_s": 0.06449364100248219,
      "per_request_mean_itl_p99_s": 0.015341997667386262,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.00503140705914121,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.8062015503876,
      "gpu_temp_max_c": 69.0,
      "gpu_power_mean_w": 39.04401550387597,
      "host_mem_available_min_bytes": 84042043392.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "refine-s2-2",
      "seqs": 2,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6679836043911467,
      "actual_dispatch_rps": 0.6679867483267269,
      "achieved_rps": 0.6662071868120987,
      "output_tokens_per_second": 85.27451991194863,
      "observed_output_tokens_per_second_all_statuses": 85.27451991194863,
      "e2e_p50_s": 2.093841230213002,
      "e2e_p95_s": 2.1243742761980684,
      "e2e_p99_s": 2.1338644631220087,
      "ttft_p99_s": 0.16267763747411665,
      "itl_p99_s": 0.051698009680112594,
      "per_request_mean_itl_p99_s": 0.015651734386225632,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0020187473426631192,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 73.0,
      "gpu_power_mean_w": 37.22850793650792,
      "host_mem_available_min_bytes": 84034904064.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "refine-s4-0",
      "seqs": 4,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.8111686066721198,
      "actual_dispatch_rps": 0.8111655437738825,
      "achieved_rps": 0.8060067761708857,
      "output_tokens_per_second": 103.16886734987337,
      "observed_output_tokens_per_second_all_statuses": 103.16886734987337,
      "e2e_p50_s": 2.3205852147948463,
      "e2e_p95_s": 2.3826821225287858,
      "e2e_p99_s": 2.4400625006912744,
      "ttft_p99_s": 0.28703043961606467,
      "itl_p99_s": 0.08970993879775056,
      "per_request_mean_itl_p99_s": 0.017697769829109183,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.00495932764955797,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 39.45088349514564,
      "host_mem_available_min_bytes": 84005150720.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "refine-s4-1",
      "seqs": 4,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.7507624338348342,
      "actual_dispatch_rps": 0.7507662875899976,
      "achieved_rps": 0.7473097636452933,
      "output_tokens_per_second": 95.65564974659755,
      "observed_output_tokens_per_second_all_statuses": 95.65564974659755,
      "e2e_p50_s": 2.2364360899518942,
      "e2e_p95_s": 2.2793786569956866,
      "e2e_p99_s": 2.2895325617199704,
      "ttft_p99_s": 0.1592414518975421,
      "itl_p99_s": 0.05119345506725959,
      "per_request_mean_itl_p99_s": 0.01689270765114668,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0022079484497953858,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.76991150442478,
      "gpu_temp_max_c": 71.0,
      "gpu_power_mean_w": 39.394530973451324,
      "host_mem_available_min_bytes": 84007772160.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "refine-s4-2",
      "seqs": 4,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.7205593474161915,
      "actual_dispatch_rps": 0.7205543421960158,
      "achieved_rps": 0.7177354060121465,
      "output_tokens_per_second": 91.87013196955475,
      "observed_output_tokens_per_second_all_statuses": 91.87013196955475,
      "e2e_p50_s": 2.1951440880511655,
      "e2e_p95_s": 2.2660098432905214,
      "e2e_p99_s": 2.3138630275665495,
      "ttft_p99_s": 0.3053598644118758,
      "itl_p99_s": 0.08678069756060725,
      "per_request_mean_itl_p99_s": 0.016689507933181125,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002288298038183721,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 93.51724137931035,
      "gpu_temp_max_c": 73.0,
      "gpu_power_mean_w": 39.18558620689655,
      "host_mem_available_min_bytes": 83851468800.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "refine-s8-0",
      "seqs": 8,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.8111686066721198,
      "actual_dispatch_rps": 0.8111673027698225,
      "achieved_rps": 0.8063047411003338,
      "output_tokens_per_second": 103.20700686084272,
      "observed_output_tokens_per_second_all_statuses": 103.20700686084272,
      "e2e_p50_s": 2.2878868812840665,
      "e2e_p95_s": 2.3385738088756627,
      "e2e_p99_s": 2.3551567183906443,
      "ttft_p99_s": 0.163683100277267,
      "itl_p99_s": 0.05327504224784214,
      "per_request_mean_itl_p99_s": 0.01739918564308314,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002126904028991703,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 72.0,
      "gpu_power_mean_w": 40.21218446601943,
      "host_mem_available_min_bytes": 83821076480.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "refine-s8-1",
      "seqs": 8,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.7507624338348342,
      "actual_dispatch_rps": 0.7507581064187059,
      "achieved_rps": 0.7471947417507949,
      "output_tokens_per_second": 95.64092694410175,
      "observed_output_tokens_per_second_all_statuses": 95.64092694410175,
      "e2e_p50_s": 2.220135329182085,
      "e2e_p95_s": 2.2639829906875093,
      "e2e_p99_s": 2.283101918855682,
      "ttft_p99_s": 0.1635813064203831,
      "itl_p99_s": 0.052227046951738916,
      "per_request_mean_itl_p99_s": 0.016790253249271803,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0021931958234927156,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 39.207687500000006,
      "host_mem_available_min_bytes": 83964940288.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "refine-s8-2",
      "seqs": 8,
      "requested": 100,
      "launched": 100,
      "completed": 100,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.7205593474161915,
      "actual_dispatch_rps": 0.7205639715804464,
      "achieved_rps": 0.7177960892977718,
      "output_tokens_per_second": 91.8778994301148,
      "observed_output_tokens_per_second_all_statuses": 91.8778994301148,
      "e2e_p50_s": 2.1714138299503247,
      "e2e_p95_s": 2.222899506950489,
      "e2e_p99_s": 2.264170159990026,
      "ttft_p99_s": 0.2991479926167815,
      "itl_p99_s": 0.07416298275522423,
      "per_request_mean_itl_p99_s": 0.016345224079992678,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0019512933639634874,
      "output_length_counts": {
        "128": 100
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.78070175438596,
      "gpu_temp_max_c": 74.0,
      "gpu_power_mean_w": 39.47835964912281,
      "host_mem_available_min_bytes": 83981275136.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "sweep-s1-0",
      "seqs": 1,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.28058985357823446,
      "actual_dispatch_rps": 0.28058994611160465,
      "achieved_rps": 0.28235754765649257,
      "output_tokens_per_second": 36.14176610003105,
      "observed_output_tokens_per_second_all_statuses": 36.14176610003105,
      "e2e_p50_s": 1.7687448280776152,
      "e2e_p95_s": 1.7947538255742983,
      "e2e_p99_s": 1.8414138916118825,
      "ttft_p99_s": 0.17266393074707578,
      "itl_p99_s": 0.0262073508329922,
      "per_request_mean_itl_p99_s": 0.01323492272489903,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.00259853416428086,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 45.38912133891213,
      "gpu_temp_max_c": 61.0,
      "gpu_power_mean_w": 24.592803347280334,
      "host_mem_available_min_bytes": 84061753344.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s1-1",
      "seqs": 1,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.37879630233061656,
      "actual_dispatch_rps": 0.37879677647642823,
      "achieved_rps": 0.3803614519060173,
      "output_tokens_per_second": 48.68626584397021,
      "observed_output_tokens_per_second_all_statuses": 48.68626584397021,
      "e2e_p50_s": 1.7704767636532779,
      "e2e_p95_s": 1.7954359504168678,
      "e2e_p99_s": 1.8594393936172127,
      "ttft_p99_s": 0.17656418296435714,
      "itl_p99_s": 0.021759753395599532,
      "per_request_mean_itl_p99_s": 0.013249812530330515,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0031281178310746185,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 69.0561797752809,
      "gpu_temp_max_c": 64.0,
      "gpu_power_mean_w": 28.509983146067416,
      "host_mem_available_min_bytes": 84086145024.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s1-2",
      "seqs": 1,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5113750081463324,
      "actual_dispatch_rps": 0.5113729250563621,
      "achieved_rps": 0.5119896251896805,
      "output_tokens_per_second": 65.5346720242791,
      "observed_output_tokens_per_second_all_statuses": 65.5346720242791,
      "e2e_p50_s": 1.7718946300665266,
      "e2e_p95_s": 1.8042855733066971,
      "e2e_p99_s": 1.8348902318779434,
      "ttft_p99_s": 0.2196709835373613,
      "itl_p99_s": 0.02413598345359783,
      "per_request_mean_itl_p99_s": 0.0132239385591703,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0020290874523925593,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 60.80916030534351,
      "gpu_temp_max_c": 68.0,
      "gpu_power_mean_w": 34.1126106870229,
      "host_mem_available_min_bytes": 84241567744.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s1-3",
      "seqs": 1,
      "requested": 80,
      "launched": 21,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 5,
      "offered_rps": 0.6903562609975488,
      "actual_dispatch_rps": 0.6903524297484306,
      "achieved_rps": 0.5493685918631593,
      "output_tokens_per_second": 70.3191797584844,
      "observed_output_tokens_per_second_all_statuses": 72.07029214504821,
      "e2e_p50_s": 4.204198197949154,
      "e2e_p95_s": 6.368799204043171,
      "e2e_p99_s": 6.554784847116388,
      "ttft_p99_s": 4.897901478817948,
      "itl_p99_s": 0.021602140825998525,
      "per_request_mean_itl_p99_s": 0.013131543999982632,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0020862816687440503,
      "output_length_counts": {
        "128": 16,
        "None": 5
      },
      "max_running": 1.0,
      "max_waiting": 4.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 69.0,
      "gpu_power_mean_w": 36.1086,
      "host_mem_available_min_bytes": 84335882240.0,
      "valid": false,
      "passes_limit": false
    },
    {
      "name": "sweep-s2-0",
      "seqs": 2,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.28058985357823446,
      "actual_dispatch_rps": 0.28058936744002294,
      "achieved_rps": 0.2823578553445338,
      "output_tokens_per_second": 36.141805484100324,
      "observed_output_tokens_per_second_all_statuses": 36.141805484100324,
      "e2e_p50_s": 1.7623595276818378,
      "e2e_p95_s": 1.787216011793498,
      "e2e_p99_s": 1.8114335150648546,
      "ttft_p99_s": 0.17601923221533053,
      "itl_p99_s": 0.02522551440430105,
      "per_request_mean_itl_p99_s": 0.013112083698376988,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0022942590864840867,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 54.62761506276151,
      "gpu_temp_max_c": 62.0,
      "gpu_power_mean_w": 25.119217573221764,
      "host_mem_available_min_bytes": 84048760832.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s2-1",
      "seqs": 2,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.37879630233061656,
      "actual_dispatch_rps": 0.37879661817034754,
      "achieved_rps": 0.38038546118298205,
      "output_tokens_per_second": 48.6893390314217,
      "observed_output_tokens_per_second_all_statuses": 48.6893390314217,
      "e2e_p50_s": 1.7753059576061787,
      "e2e_p95_s": 1.8357785279360541,
      "e2e_p99_s": 1.867389821490069,
      "ttft_p99_s": 0.2647945922701909,
      "itl_p99_s": 0.026049931525485577,
      "per_request_mean_itl_p99_s": 0.013280917272227449,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0024709164866362674,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 57.27272727272727,
      "gpu_temp_max_c": 63.0,
      "gpu_power_mean_w": 27.3621875,
      "host_mem_available_min_bytes": 84059508736.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s2-2",
      "seqs": 2,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5113750081463324,
      "actual_dispatch_rps": 0.5113759017208952,
      "achieved_rps": 0.5118009852026001,
      "output_tokens_per_second": 65.51052610593281,
      "observed_output_tokens_per_second_all_statuses": 65.51052610593281,
      "e2e_p50_s": 1.7649004480917938,
      "e2e_p95_s": 1.8070464312018886,
      "e2e_p99_s": 1.8611183810453802,
      "ttft_p99_s": 0.31509359978095725,
      "itl_p99_s": 0.02328377307232584,
      "per_request_mean_itl_p99_s": 0.013370167114543885,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002182120979850878,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 68.0,
      "gpu_power_mean_w": 33.33161538461539,
      "host_mem_available_min_bytes": 84052353024.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s2-3",
      "seqs": 2,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6903562609975488,
      "actual_dispatch_rps": 0.6903540090625729,
      "achieved_rps": 0.6874955371769514,
      "output_tokens_per_second": 87.99942875864978,
      "observed_output_tokens_per_second_all_statuses": 87.99942875864978,
      "e2e_p50_s": 2.1306447907481925,
      "e2e_p95_s": 2.185290587609052,
      "e2e_p99_s": 2.3373866072234515,
      "ttft_p99_s": 0.2595948062192454,
      "itl_p99_s": 0.06103171747570739,
      "per_request_mean_itl_p99_s": 0.01631793266503616,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0017141179420286758,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 37.93177319587628,
      "host_mem_available_min_bytes": 84103581696.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "sweep-s4-0",
      "seqs": 4,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.28058985357823446,
      "actual_dispatch_rps": 0.2805888784244751,
      "achieved_rps": 0.28227588475691645,
      "output_tokens_per_second": 36.131313248885306,
      "observed_output_tokens_per_second_all_statuses": 36.131313248885306,
      "e2e_p50_s": 1.7703593843471026,
      "e2e_p95_s": 1.7940466236439534,
      "e2e_p99_s": 1.8636019138008122,
      "ttft_p99_s": 0.17799579055179438,
      "itl_p99_s": 0.027648425986408223,
      "per_request_mean_itl_p99_s": 0.013649830096231693,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0028146227831894045,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 44.773109243697476,
      "gpu_temp_max_c": 60.0,
      "gpu_power_mean_w": 23.89522268907563,
      "host_mem_available_min_bytes": 83898712064.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s4-1",
      "seqs": 4,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.37879630233061656,
      "actual_dispatch_rps": 0.3787961198647226,
      "achieved_rps": 0.3803325809865261,
      "output_tokens_per_second": 48.68257036627534,
      "observed_output_tokens_per_second_all_statuses": 48.68257036627534,
      "e2e_p50_s": 1.7667437709824299,
      "e2e_p95_s": 1.8076212854684854,
      "e2e_p99_s": 1.8270793029855121,
      "ttft_p99_s": 0.2100283748685612,
      "itl_p99_s": 0.028850893003982428,
      "per_request_mean_itl_p99_s": 0.013328805786455976,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.003171155113668638,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 55.32203389830509,
      "gpu_temp_max_c": 63.0,
      "gpu_power_mean_w": 28.01746327683616,
      "host_mem_available_min_bytes": 84074778624.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s4-2",
      "seqs": 4,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5113750081463324,
      "actual_dispatch_rps": 0.5113753941495022,
      "achieved_rps": 0.5120033930178428,
      "output_tokens_per_second": 65.53643430628388,
      "observed_output_tokens_per_second_all_statuses": 65.53643430628388,
      "e2e_p50_s": 1.771713177447964,
      "e2e_p95_s": 1.8033918680404895,
      "e2e_p99_s": 1.854976584124815,
      "ttft_p99_s": 0.2414804889744838,
      "itl_p99_s": 0.03131132682610768,
      "per_request_mean_itl_p99_s": 0.013147012141030638,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.001791383285308254,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.8,
      "gpu_temp_max_c": 67.0,
      "gpu_power_mean_w": 34.34946923076924,
      "host_mem_available_min_bytes": 83938754560.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s4-3",
      "seqs": 4,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6903562609975488,
      "actual_dispatch_rps": 0.6903544932409008,
      "achieved_rps": 0.6875938370782885,
      "output_tokens_per_second": 88.01201114602092,
      "observed_output_tokens_per_second_all_statuses": 88.01201114602092,
      "e2e_p50_s": 2.1396440300813993,
      "e2e_p95_s": 2.177405054226983,
      "e2e_p99_s": 2.185582915441628,
      "ttft_p99_s": 0.18026874369723372,
      "itl_p99_s": 0.053480047331540805,
      "per_request_mean_itl_p99_s": 0.016050770391626917,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0020642399272765032,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 38.57863917525773,
      "host_mem_available_min_bytes": 83917856768.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s4-4",
      "seqs": 4,
      "requested": 83,
      "launched": 83,
      "completed": 83,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.9319809523466909,
      "actual_dispatch_rps": 0.9319794637662836,
      "achieved_rps": 0.9204296194821155,
      "output_tokens_per_second": 117.81499129371079,
      "observed_output_tokens_per_second_all_statuses": 117.81499129371079,
      "e2e_p50_s": 2.7440112509793835,
      "e2e_p95_s": 2.8180025757945257,
      "e2e_p99_s": 2.8373681125839356,
      "ttft_p99_s": 0.17510360672051298,
      "itl_p99_s": 0.0640173873951426,
      "per_request_mean_itl_p99_s": 0.02120077991070259,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0025547526185982702,
      "output_length_counts": {
        "128": 83
      },
      "max_running": 3.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.12631578947368416,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 70.0,
      "gpu_power_mean_w": 41.25338666666668,
      "host_mem_available_min_bytes": 83963166720.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "sweep-s8-0",
      "seqs": 8,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.28058985357823446,
      "actual_dispatch_rps": 0.2805966053964548,
      "achieved_rps": 0.2823769624013062,
      "output_tokens_per_second": 36.144251187367196,
      "observed_output_tokens_per_second_all_statuses": 36.144251187367196,
      "e2e_p50_s": 1.7642738126014592,
      "e2e_p95_s": 1.7878187624795827,
      "e2e_p99_s": 1.8605267704922879,
      "ttft_p99_s": 0.20614301679132027,
      "itl_p99_s": 0.022038180015661053,
      "per_request_mean_itl_p99_s": 0.013189405380821773,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.003328161441895613,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 44.962025316455694,
      "gpu_temp_max_c": 60.0,
      "gpu_power_mean_w": 24.789818565400843,
      "host_mem_available_min_bytes": 83857395712.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s8-1",
      "seqs": 8,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.37879630233061656,
      "actual_dispatch_rps": 0.3787953307214756,
      "achieved_rps": 0.3803421256684687,
      "output_tokens_per_second": 48.683792085563994,
      "observed_output_tokens_per_second_all_statuses": 48.683792085563994,
      "e2e_p50_s": 1.766815075701743,
      "e2e_p95_s": 1.7972581824767986,
      "e2e_p99_s": 1.8316668186314926,
      "ttft_p99_s": 0.1819343460854723,
      "itl_p99_s": 0.02258395183729589,
      "per_request_mean_itl_p99_s": 0.01320766423118835,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002984692358440931,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 68.72727272727273,
      "gpu_temp_max_c": 63.0,
      "gpu_power_mean_w": 29.762295454545452,
      "host_mem_available_min_bytes": 83819749376.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s8-2",
      "seqs": 8,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5113750081463324,
      "actual_dispatch_rps": 0.51137652162495,
      "achieved_rps": 0.5119913808110231,
      "output_tokens_per_second": 65.53489674381096,
      "observed_output_tokens_per_second_all_statuses": 65.53489674381096,
      "e2e_p50_s": 1.7688242646181607,
      "e2e_p95_s": 1.844617390450003,
      "e2e_p99_s": 1.8660939411916477,
      "ttft_p99_s": 0.23684129552449973,
      "itl_p99_s": 0.029437468241958406,
      "per_request_mean_itl_p99_s": 0.01336540029714994,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.001962191958446054,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 94.13953488372093,
      "gpu_temp_max_c": 69.0,
      "gpu_power_mean_w": 34.82882945736434,
      "host_mem_available_min_bytes": 83921227776.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s8-3",
      "seqs": 8,
      "requested": 80,
      "launched": 80,
      "completed": 80,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.6903562609975488,
      "actual_dispatch_rps": 0.6903614975521761,
      "achieved_rps": 0.6875274009786412,
      "output_tokens_per_second": 88.00350732526607,
      "observed_output_tokens_per_second_all_statuses": 88.00350732526607,
      "e2e_p50_s": 2.1226487555759377,
      "e2e_p95_s": 2.1477595713316986,
      "e2e_p99_s": 2.15272330380365,
      "ttft_p99_s": 0.16471970153463186,
      "itl_p99_s": 0.05174143187381558,
      "per_request_mean_itl_p99_s": 0.015896796450643583,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002666545369138469,
      "output_length_counts": {
        "128": 80
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 95.91752577319588,
      "gpu_temp_max_c": 68.0,
      "gpu_power_mean_w": 38.16879381443299,
      "host_mem_available_min_bytes": 83807666176.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "sweep-s8-4",
      "seqs": 8,
      "requested": 83,
      "launched": 83,
      "completed": 83,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.9319809523466909,
      "actual_dispatch_rps": 0.9319701238410572,
      "achieved_rps": 0.9206876932005725,
      "output_tokens_per_second": 117.84802472967328,
      "observed_output_tokens_per_second_all_statuses": 117.84802472967328,
      "e2e_p50_s": 2.6868563337338855,
      "e2e_p95_s": 2.774293575732736,
      "e2e_p99_s": 2.8117018926632587,
      "ttft_p99_s": 0.2878068665994214,
      "itl_p99_s": 0.10886386620113696,
      "per_request_mean_itl_p99_s": 0.020872491200738235,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002066624333383514,
      "output_length_counts": {
        "128": 83
      },
      "max_running": 3.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.12631578947368416,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 96.0,
      "gpu_temp_max_c": 69.0,
      "gpu_power_mean_w": 40.80202666666667,
      "host_mem_available_min_bytes": 83828989952.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "threshold-baseline-s1",
      "seqs": 1,
      "requested": 1000,
      "launched": 1000,
      "completed": 1000,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.28058985357823446,
      "actual_dispatch_rps": 0.2805898475981577,
      "achieved_rps": 0.28073150425500265,
      "output_tokens_per_second": 35.93363254464034,
      "observed_output_tokens_per_second_all_statuses": 35.93363254464034,
      "e2e_p50_s": 1.769770291444729,
      "e2e_p95_s": 1.8074064880354854,
      "e2e_p99_s": 1.883460009765986,
      "ttft_p99_s": 0.22975969112623715,
      "itl_p99_s": 0.026637616686348307,
      "per_request_mean_itl_p99_s": 0.013335607410593033,
      "coalesced_events": 0,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002640927867614664,
      "output_length_counts": {
        "128": 1000
      },
      "max_running": 1.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 45.49165554072096,
      "gpu_temp_max_c": 62.0,
      "gpu_power_mean_w": 24.351521695594126,
      "host_mem_available_min_bytes": 83840266240.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "warm-1-0",
      "seqs": 1,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.49999931835089995,
      "achieved_rps": 0.5002342522436214,
      "output_tokens_per_second": 64.02998428718354,
      "observed_output_tokens_per_second_all_statuses": 64.02998428718354,
      "e2e_p50_s": 1.7880326590020559,
      "e2e_p95_s": 2.0700507650035433,
      "e2e_p99_s": 2.204720304196235,
      "ttft_p99_s": 0.9766665831499265,
      "itl_p99_s": 0.03040933598007545,
      "per_request_mean_itl_p99_s": 0.013110899177154637,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0019855308950354807,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 1.0,
      "max_waiting": 1.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 73.92857142857143,
      "gpu_temp_max_c": 52.0,
      "gpu_power_mean_w": 30.930500000000002,
      "host_mem_available_min_bytes": 83984109568.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "warm-1-36",
      "seqs": 1,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.4999944282288215,
      "achieved_rps": 0.5037946293233293,
      "output_tokens_per_second": 64.48571255338615,
      "observed_output_tokens_per_second_all_statuses": 64.48571255338615,
      "e2e_p50_s": 1.7644154320078087,
      "e2e_p95_s": 2.069147469504969,
      "e2e_p99_s": 2.194729606708279,
      "ttft_p99_s": 0.9396818063585667,
      "itl_p99_s": 0.02079916818765925,
      "per_request_mean_itl_p99_s": 0.013142563570836215,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.003930777703499188,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 1.0,
      "max_waiting": 1.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 0.0,
      "gpu_temp_max_c": 57.0,
      "gpu_power_mean_w": 23.817259259259256,
      "host_mem_available_min_bytes": 83922980864.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "warm-1-43",
      "seqs": 1,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.49999444122848563,
      "achieved_rps": 0.5032195089868741,
      "output_tokens_per_second": 64.41209715031988,
      "observed_output_tokens_per_second_all_statuses": 64.41209715031988,
      "e2e_p50_s": 1.7766348960067262,
      "e2e_p95_s": 2.013890348509449,
      "e2e_p99_s": 2.153986648904538,
      "ttft_p99_s": 0.894564542653097,
      "itl_p99_s": 0.020354802655638195,
      "per_request_mean_itl_p99_s": 0.013129569881462298,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0020017503084091,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 1.0,
      "max_waiting": 1.0,
      "kv_cache_use_max": 0.04210526315789476,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 87.66666666666667,
      "gpu_temp_max_c": 57.0,
      "gpu_power_mean_w": 34.134,
      "host_mem_available_min_bytes": 83829669888.0,
      "valid": true,
      "passes_limit": true
    },
    {
      "name": "warm-2-10",
      "seqs": 2,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.4999924758133566,
      "achieved_rps": 0.503854723192538,
      "output_tokens_per_second": 64.49340456864486,
      "observed_output_tokens_per_second_all_statuses": 64.49340456864486,
      "e2e_p50_s": 1.7644736600050237,
      "e2e_p95_s": 2.026544375512458,
      "e2e_p99_s": 2.3098546879118658,
      "ttft_p99_s": 0.9651290557543686,
      "itl_p99_s": 0.02186259337569937,
      "per_request_mean_itl_p99_s": 0.013220473650776905,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0021687288040993734,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 49.77777777777778,
      "gpu_temp_max_c": 57.0,
      "gpu_power_mean_w": 24.488925925925926,
      "host_mem_available_min_bytes": 83947728896.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "warm-4-18",
      "seqs": 4,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.4999855316518809,
      "achieved_rps": 0.5035818488050187,
      "output_tokens_per_second": 64.45847664704239,
      "observed_output_tokens_per_second_all_statuses": 64.45847664704239,
      "e2e_p50_s": 1.776132058505027,
      "e2e_p95_s": 2.0294500922536827,
      "e2e_p99_s": 2.325693610454618,
      "ttft_p99_s": 0.9051100200042126,
      "itl_p99_s": 0.0964957253122704,
      "per_request_mean_itl_p99_s": 0.013372122173210859,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.001959977901424281,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 73.84615384615384,
      "gpu_temp_max_c": 54.0,
      "gpu_power_mean_w": 26.73538461538461,
      "host_mem_available_min_bytes": 84010991616.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "warm-4-38",
      "seqs": 4,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.5000062814121495,
      "achieved_rps": 0.5030270272128183,
      "output_tokens_per_second": 64.38745948324075,
      "observed_output_tokens_per_second_all_statuses": 64.38745948324075,
      "e2e_p50_s": 1.812586663516413,
      "e2e_p95_s": 2.0712814565122244,
      "e2e_p99_s": 2.3731956425137466,
      "ttft_p99_s": 0.9379809957092218,
      "itl_p99_s": 0.02373792076614337,
      "per_request_mean_itl_p99_s": 0.013619112001555994,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0018204911619250197,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 10.666666666666666,
      "gpu_temp_max_c": 59.0,
      "gpu_power_mean_w": 15.368666666666664,
      "host_mem_available_min_bytes": 84056510464.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "warm-4-40",
      "seqs": 4,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.499993994521941,
      "achieved_rps": 0.5033477576467116,
      "output_tokens_per_second": 64.42851297877908,
      "observed_output_tokens_per_second_all_statuses": 64.42851297877908,
      "e2e_p50_s": 1.7689110924984561,
      "e2e_p95_s": 2.0217522679995454,
      "e2e_p99_s": 2.331738991996826,
      "ttft_p99_s": 0.9119643580452245,
      "itl_p99_s": 0.025097039240645252,
      "per_request_mean_itl_p99_s": 0.013339175094499053,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.0020898742419376503,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 88.66666666666667,
      "gpu_temp_max_c": 57.0,
      "gpu_power_mean_w": 35.669111111111114,
      "host_mem_available_min_bytes": 83840323584.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "warm-4-46",
      "seqs": 4,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.5000488881128965,
      "achieved_rps": 0.5028594326288225,
      "output_tokens_per_second": 64.36600737648928,
      "observed_output_tokens_per_second_all_statuses": 64.36600737648928,
      "e2e_p50_s": 1.8007311395122088,
      "e2e_p95_s": 2.0458888550128904,
      "e2e_p99_s": 2.3419102118132287,
      "ttft_p99_s": 0.9663840271605292,
      "itl_p99_s": 0.022128592839872008,
      "per_request_mean_itl_p99_s": 0.01356251034879343,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.003593529012141516,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 42.666666666666664,
      "gpu_temp_max_c": 58.0,
      "gpu_power_mean_w": 23.044222222222217,
      "host_mem_available_min_bytes": 83954622464.0,
      "valid": true,
      "passes_limit": false
    },
    {
      "name": "warm-8-27",
      "seqs": 8,
      "requested": 16,
      "launched": 16,
      "completed": 16,
      "errors": 0,
      "timeouts": 0,
      "cancelled": 0,
      "offered_rps": 0.5,
      "actual_dispatch_rps": 0.5000128946158153,
      "achieved_rps": 0.5033305644393016,
      "output_tokens_per_second": 64.4263122482306,
      "observed_output_tokens_per_second_all_statuses": 64.4263122482306,
      "e2e_p50_s": 1.760774307003885,
      "e2e_p95_s": 1.9907277790043736,
      "e2e_p99_s": 2.248330763800186,
      "ttft_p99_s": 0.8454717812484884,
      "itl_p99_s": 0.031214583390392483,
      "per_request_mean_itl_p99_s": 0.013217271020019282,
      "coalesced_events": 1,
      "length_invalid": 0,
      "token_timing_invalid": 0,
      "dispatch_lag_p99_s": 0.002198200758721214,
      "output_length_counts": {
        "128": 16
      },
      "max_running": 2.0,
      "max_waiting": 0.0,
      "kv_cache_use_max": 0.08421052631578951,
      "preemptions_delta": 0.0,
      "gpu_util_mean_pct": 66.46153846153847,
      "gpu_temp_max_c": 57.0,
      "gpu_power_mean_w": 26.861538461538462,
      "host_mem_available_min_bytes": 83654410240.0,
      "valid": true,
      "passes_limit": true
    }
  ],
  "restoration": {
    "verified": true,
    "restored_at_epoch": 1792098374.9218853,
    "original_configuration_restored": true,
    "routing_selectors_unchanged": true,
    "qwen_generation_verified": true,
    "muse_model_list_verified": true
  },
  "evidence_scope": "Public aggregates and benchmark methodology only. Raw per-request records, telemetry, deployment configuration, workload and scripts are retained in the local experiment artifacts."
}
