{
  "caveats": [
    "Bounded lab package proof only; configured 262144 context is not long-context quality. Memory is sampled whole-host unavailable memory during speed, not allocator residency or the entire drive peak. Candidate: one concurrency-2 answer returned fenced JSON and failed the unchanged strict contract. Uncalibrated FP8 KV scales remain a quality caveat.",
    "Weights file SHA-256 was not recorded for public reproduction."
  ],
  "engine": {
    "name": "SGLang",
    "recipe": "speculative decoding",
    "version": "0.0.0.dev0+qwen38.27b.g561c8f3"
  },
  "format": "olir/0.1",
  "history": {
    "corrections": [],
    "supersedes": []
  },
  "id": "4d30da78f0b56f97709f43d6",
  "measured_at": "2026-10-05T22:21:12.790316+00:00",
  "metrics": [
    {
      "basis": null,
      "n": null,
      "name": "workload_total",
      "scope": "machine_total",
      "source_field": "workload_total",
      "statistic": "single",
      "unit": "count",
      "value": 16
    },
    {
      "basis": null,
      "n": 3,
      "name": "chat_decode_tps",
      "samples": [
        34.347989734983216,
        34.369737142787336,
        34.28250445000213
      ],
      "scope": "per_person",
      "source_field": "chat_decode_tps",
      "statistic": "mean",
      "stddev": 0.045407075413807445,
      "unit": "tok/s",
      "value": 34.333410442590896
    },
    {
      "basis": null,
      "n": null,
      "name": "requests_ok",
      "scope": "machine_total",
      "source_field": "requests_ok",
      "statistic": "single",
      "unit": "count",
      "value": 6
    },
    {
      "basis": null,
      "n": 3,
      "name": "code_ttft_mean_s",
      "samples": [
        0.19521947705652565,
        0.19531845301389694,
        0.19371042301645502
      ],
      "scope": "per_person",
      "source_field": "code_ttft_mean_s",
      "statistic": "mean",
      "stddev": 0.0009011844806201899,
      "unit": "s",
      "value": 0.19474945102895921
    },
    {
      "basis": null,
      "n": null,
      "name": "workload_passed",
      "scope": "machine_total",
      "source_field": "workload_passed",
      "statistic": "single",
      "unit": "count",
      "value": 15
    },
    {
      "basis": null,
      "n": null,
      "name": "concurrency",
      "scope": "machine_total",
      "source_field": "workload_levels[0].concurrency",
      "statistic": "single",
      "unit": "count",
      "value": 1
    },
    {
      "basis": null,
      "n": null,
      "name": "passed",
      "scope": "machine_total",
      "source_field": "workload_levels[0].passed",
      "statistic": "single",
      "unit": "count",
      "value": 16
    },
    {
      "basis": null,
      "n": null,
      "name": "total",
      "scope": "machine_total",
      "source_field": "workload_levels[0].total",
      "statistic": "single",
      "unit": "count",
      "value": 16
    },
    {
      "basis": null,
      "n": null,
      "name": "concurrency",
      "scope": "machine_total",
      "source_field": "workload_levels[1].concurrency",
      "statistic": "single",
      "unit": "count",
      "value": 2
    },
    {
      "basis": null,
      "n": null,
      "name": "passed",
      "scope": "machine_total",
      "source_field": "workload_levels[1].passed",
      "statistic": "single",
      "unit": "count",
      "value": 15
    },
    {
      "basis": null,
      "n": null,
      "name": "total",
      "scope": "machine_total",
      "source_field": "workload_levels[1].total",
      "statistic": "single",
      "unit": "count",
      "value": 16
    },
    {
      "basis": null,
      "n": null,
      "name": "aggregate_concurrency",
      "scope": "machine_total",
      "source_field": "aggregate_concurrency",
      "statistic": "single",
      "unit": "count",
      "value": 2
    },
    {
      "basis": null,
      "n": null,
      "name": "requests_total",
      "scope": "machine_total",
      "source_field": "requests_total",
      "statistic": "single",
      "unit": "count",
      "value": 6
    },
    {
      "basis": null,
      "n": null,
      "name": "aggregate_e2e_tps",
      "scope": "machine_total",
      "source_field": "aggregate_tps",
      "statistic": "single",
      "unit": "tok/s",
      "value": 56.92073838049442
    },
    {
      "basis": null,
      "n": 3,
      "name": "code_decode_tps",
      "samples": [
        57.62747328933936,
        57.454320212510794,
        57.50305099811821
      ],
      "scope": "per_person",
      "source_field": "code_decode_tps",
      "statistic": "mean",
      "stddev": 0.08929126873500154,
      "unit": "tok/s",
      "value": 57.528281499989454
    },
    {
      "basis": null,
      "n": 3,
      "name": "chat_ttft_mean_s",
      "samples": [
        0.1922413359861821,
        0.19155225198483095,
        0.1926738429465331
      ],
      "scope": "per_person",
      "source_field": "chat_ttft_mean_s",
      "statistic": "mean",
      "stddev": 0.0005656655839277865,
      "unit": "s",
      "value": 0.19215581030584872
    },
    {
      "basis": null,
      "n": null,
      "name": "peak_host_unavailable_gib",
      "scope": "machine_total",
      "source_field": "peak_host_unavailable_gib",
      "statistic": "max",
      "unit": "GiB",
      "value": 113.64101028442383
    }
  ],
  "model": {
    "artifact_sha256": [],
    "publisher_repo": "Qwen/Qwen3.8-27B",
    "publisher_revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
    "variant": "NVFP4 variant",
    "weights_dtype": "NVFP4",
    "weights_format": "safetensors"
  },
  "outcome": {
    "gates": [
      {
        "met": true,
        "name": "requests",
        "observed": "6/6"
      },
      {
        "met": false,
        "name": "structured-tasks",
        "observed": "15/16"
      }
    ],
    "reasons": [
      "15/16 tasks passed."
    ],
    "status": "partial"
  },
  "publication_group": "pre-standard",
  "published": null,
  "publisher": {
    "name": "Skilak Keep"
  },
  "quality": [
    {
      "passed": 15,
      "skipped": null,
      "suite": "structured-tasks",
      "total": 16,
      "version": null
    }
  ],
  "record_date": "2026-10-05",
  "reproduce": {
    "driver": null,
    "fixture_sha256": null,
    "harness": {
      "name": null,
      "version": null
    },
    "harness_sha256": "446eae1b7d949003bd1379b1d1614c10b4f484a093cea9a3dbab7db56672fe5b",
    "os": null,
    "workload_name": null,
    "workload_version": null
  },
  "requests": [],
  "scenario": {
    "id": "workload-qualification",
    "version": null
  },
  "settings": {
    "concurrency": 1,
    "context_tokens": 262144,
    "kv_dtype": "fp8_e4m3",
    "max_output_tokens": 512,
    "speculative_decoding": "speculative decoding"
  },
  "setup_complete": false,
  "setup_id": "bcdf385cd1ff43923b9a7bfaf9fb1df0",
  "standard": "pre-standard",
  "system": {
    "driver": null,
    "hardware": "NVIDIA GB10, 128 GB unified memory",
    "os": null
  },
  "workload": {
    "cache": null,
    "content_tokens": null,
    "document": null,
    "input_tokens": null,
    "output_tokens": 512,
    "repetitions": null,
    "users": 1,
    "warmup": null
  }
}