{
  "caveats": [
    "Short repeated code/chat prompts; output hit the token ceiling; task correctness and full context quality unassessed. Missing FP8 KV scales defaulted to 1.0; accuracy unassessed.",
    "Engine version was not recorded for public reproduction.",
    "Weights file SHA-256 was not recorded for public reproduction."
  ],
  "engine": {
    "name": "SGLang",
    "recipe": "speculative decoding",
    "version": null
  },
  "format": "olir/0.1",
  "history": {
    "corrections": [],
    "supersedes": []
  },
  "id": "d43d3d2921d854178c4193f4",
  "measured_at": null,
  "metrics": [
    {
      "basis": null,
      "n": null,
      "name": "chat_decode_tps",
      "scope": "per_person",
      "source_field": "chat_decode_tps",
      "statistic": "mean",
      "unit": "tok/s",
      "value": 21.238341518098036
    },
    {
      "basis": null,
      "n": null,
      "name": "requests_ok",
      "scope": "machine_total",
      "source_field": "requests_ok",
      "statistic": "single",
      "unit": "count",
      "value": 6
    },
    {
      "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": "code_decode_tps",
      "scope": "per_person",
      "source_field": "code_decode_tps",
      "statistic": "mean",
      "unit": "tok/s",
      "value": 31.991125012719905
    },
    {
      "basis": null,
      "n": null,
      "name": "peak_host_unavailable_gib",
      "scope": "machine_total",
      "source_field": "peak_host_unavailable_gib",
      "statistic": "max",
      "unit": "GiB",
      "value": 115.10027694702148
    }
  ],
  "model": {
    "artifact_sha256": [],
    "publisher_repo": "Qwen/Qwen3.8-27B",
    "publisher_revision": "1d4bf0f2ff6012fd82039f2fa52739d0dd7c60c0",
    "variant": "NVFP4 variant with BF16 head",
    "weights_dtype": "NVFP4",
    "weights_format": "safetensors"
  },
  "outcome": {
    "gates": [
      {
        "met": true,
        "name": "requests",
        "observed": "6/6"
      }
    ],
    "reasons": [],
    "status": "not_graded"
  },
  "publication_group": "pre-standard",
  "published": null,
  "publisher": {
    "name": "Skilak Keep"
  },
  "quality": [],
  "record_date": "2026-10-05",
  "reproduce": {
    "driver": null,
    "fixture_sha256": null,
    "harness": {
      "name": null,
      "version": null
    },
    "harness_sha256": null,
    "os": null,
    "workload_name": null,
    "workload_version": null
  },
  "requests": [],
  "scenario": {
    "id": "llm-throughput",
    "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": "ae7164bf7eef4373dec8f83258848782",
  "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
  }
}