{
  "caveats": [
    "Lab observation, not customer acceptance. Prompt content length excludes the server chat template; actual per-request usage is recorded. Synthesis is requested and retained but semantic accuracy is unassessed. All requested users must overlap in observed decode to pass.",
    "Engine version was not recorded for public reproduction.",
    "Weights file SHA-256 was not recorded for public reproduction."
  ],
  "engine": {
    "name": "llama.cpp CUDA",
    "recipe": "plain",
    "version": null
  },
  "format": "olir/0.1",
  "history": {
    "corrections": [],
    "supersedes": []
  },
  "id": "6fc815a108f34202abcaff2e",
  "measured_at": "2026-10-07T00:56:03.723718+00:00",
  "metrics": [
    {
      "basis": null,
      "n": null,
      "name": "wall_seconds",
      "scope": "machine_total",
      "source_field": "wall_seconds",
      "statistic": "single",
      "unit": "s",
      "value": 120.90144624200184
    },
    {
      "basis": null,
      "n": 1,
      "name": "ttft_median_s",
      "scope": "per_person",
      "source_field": "ttft_median_s",
      "statistic": "median",
      "unit": "s",
      "value": 118.03834720898885
    },
    {
      "basis": null,
      "n": null,
      "name": "needle_total",
      "scope": "machine_total",
      "source_field": "needle_total",
      "statistic": "single",
      "unit": "count",
      "value": 3
    },
    {
      "basis": null,
      "n": 1,
      "name": "ttft_max_s",
      "scope": "per_person",
      "source_field": "ttft_max_s",
      "statistic": "max",
      "unit": "s",
      "value": 118.03834720898885
    },
    {
      "basis": null,
      "n": null,
      "name": "requests_ok",
      "scope": "machine_total",
      "source_field": "requests_ok",
      "statistic": "single",
      "unit": "count",
      "value": 1
    },
    {
      "basis": null,
      "n": null,
      "name": "needle_correct",
      "scope": "machine_total",
      "source_field": "needle_correct",
      "statistic": "single",
      "unit": "count",
      "value": 3
    },
    {
      "basis": null,
      "n": null,
      "name": "latency_bar_seconds",
      "scope": "machine_total",
      "source_field": "latency_bar_seconds",
      "statistic": "single",
      "unit": "s",
      "value": 300
    },
    {
      "basis": null,
      "n": null,
      "name": "decode_overlap",
      "scope": "machine_total",
      "source_field": "decode_overlap",
      "statistic": "single",
      "unit": "count",
      "value": 1
    },
    {
      "basis": null,
      "n": null,
      "name": "aggregate_concurrency",
      "scope": "machine_total",
      "source_field": "aggregate_concurrency",
      "statistic": "single",
      "unit": "count",
      "value": 1
    },
    {
      "basis": null,
      "n": null,
      "name": "ttft_s",
      "scope": "per_person",
      "source_field": "long_context_requests[0].ttft_s",
      "statistic": "single",
      "unit": "s",
      "value": 118.03834720898885
    },
    {
      "basis": null,
      "n": null,
      "name": "total_s",
      "scope": "per_person",
      "source_field": "long_context_requests[0].total_s",
      "statistic": "single",
      "unit": "s",
      "value": 120.89322150498629
    },
    {
      "basis": null,
      "n": null,
      "name": "decode_tps",
      "scope": "per_person",
      "source_field": "long_context_requests[0].decode_tps",
      "statistic": "single",
      "unit": "tok/s",
      "value": 31.174752641396164
    },
    {
      "basis": "effective, includes queueing",
      "n": null,
      "name": "prefill_tps",
      "scope": "per_person",
      "source_field": "long_context_requests[0].effective_prefill_tps",
      "statistic": "single",
      "unit": "tok/s",
      "value": 2118.0913314326785
    },
    {
      "basis": null,
      "n": null,
      "name": "user_number",
      "scope": "per_person",
      "source_field": "long_context_requests[0].user_number",
      "statistic": "single",
      "unit": "count",
      "value": 0
    },
    {
      "basis": null,
      "n": null,
      "name": "prompt_tokens",
      "scope": "per_person",
      "source_field": "long_context_requests[0].prompt_tokens",
      "statistic": "single",
      "unit": "tokens",
      "value": 250016
    },
    {
      "basis": null,
      "n": null,
      "name": "output_tokens",
      "scope": "per_person",
      "source_field": "long_context_requests[0].output_tokens",
      "statistic": "single",
      "unit": "tokens",
      "value": 90
    },
    {
      "basis": null,
      "n": null,
      "name": "needle_correct",
      "scope": "per_person",
      "source_field": "long_context_requests[0].needle_correct",
      "statistic": "single",
      "unit": "count",
      "value": 3
    },
    {
      "basis": null,
      "n": null,
      "name": "needle_total",
      "scope": "per_person",
      "source_field": "long_context_requests[0].needle_total",
      "statistic": "single",
      "unit": "count",
      "value": 3
    },
    {
      "basis": null,
      "n": null,
      "name": "first_real_token_from_dispatch_s",
      "scope": "per_person",
      "source_field": "long_context_requests[0].first_real_token_from_dispatch_s",
      "statistic": "single",
      "unit": "s",
      "value": 118.03743578796275
    },
    {
      "basis": null,
      "n": null,
      "name": "cached_prompt_tokens",
      "scope": "per_person",
      "source_field": "long_context_requests[0].cached_prompt_tokens",
      "statistic": "single",
      "unit": "count",
      "value": 0
    },
    {
      "basis": null,
      "n": null,
      "name": "requests_total",
      "scope": "machine_total",
      "source_field": "requests_total",
      "statistic": "single",
      "unit": "count",
      "value": 1
    },
    {
      "basis": null,
      "n": null,
      "name": "aggregate_decode_tps",
      "scope": "machine_total",
      "source_field": "aggregate_tps",
      "statistic": "single",
      "unit": "tok/s",
      "value": 31.174738788744083
    },
    {
      "basis": null,
      "n": null,
      "name": "aggregate_e2e_tps",
      "scope": "machine_total",
      "source_field": "aggregate_end_to_end_tps",
      "statistic": "single",
      "unit": "tok/s",
      "value": 0.7444079686181082
    },
    {
      "basis": null,
      "n": null,
      "name": "peak_host_unavailable_gib",
      "scope": "machine_total",
      "source_field": "peak_host_unavailable_gib",
      "statistic": "max",
      "unit": "GiB",
      "value": 35.79835510253906
    }
  ],
  "model": {
    "artifact_sha256": [],
    "publisher_repo": "nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16",
    "publisher_revision": "a9904d24bcc1d289a1950fa9d2b978c47cf903b9",
    "variant": "Q4_K_M GGUF variant",
    "weights_dtype": "Q4_K_M",
    "weights_format": "GGUF"
  },
  "outcome": {
    "gates": [
      {
        "met": true,
        "name": "requests",
        "observed": "1/1"
      },
      {
        "met": true,
        "name": "hidden-facts",
        "observed": "3/3"
      },
      {
        "met": true,
        "name": "first-token-300-s-reference",
        "observed": 118.03834720898885
      }
    ],
    "reasons": [],
    "status": "pass"
  },
  "publication_group": "pre-standard",
  "published": null,
  "publisher": {
    "name": "Skilak Keep"
  },
  "quality": [
    {
      "passed": 3,
      "skipped": null,
      "suite": "hidden-facts",
      "total": 3,
      "version": "1"
    }
  ],
  "record_date": "2026-10-07",
  "reproduce": {
    "driver": null,
    "fixture_sha256": "04cf656671acc024178598a26462c2de38b6f97137ea5abf44a09b83b8047c5e",
    "harness": {
      "name": "keep-long-context-concurrency-v1",
      "version": "1"
    },
    "harness_sha256": null,
    "os": null,
    "workload_name": null,
    "workload_version": null
  },
  "requests": [
    {
      "cached_prompt_tokens": 0,
      "checks_passed": true,
      "completed": true,
      "decode_tps": 31.174752641396164,
      "effective_prefill_tps": 2118.0913314326785,
      "first_real_token_from_dispatch_s": 118.03743578796275,
      "first_token_within_300_seconds": true,
      "needle_correct": 3,
      "needle_total": 3,
      "output_tokens": 90,
      "prompt_tokens": 250016,
      "stop_and_done": true,
      "strict_json": true,
      "total_s": 120.89322150498629,
      "ttft_s": 118.03834720898885,
      "user_number": 0,
      "valid_json": true
    }
  ],
  "scenario": {
    "id": "long-context-concurrency",
    "version": "1"
  },
  "settings": {
    "cache_state": "cold",
    "concurrency": 1,
    "configured_context_tokens": 262144,
    "context_tokens": 250000,
    "fixture_variant": "distinct-v1",
    "kv_dtype": "q8_0",
    "max_output_tokens": 512,
    "scenario": "distinct",
    "thinking": false
  },
  "setup_complete": false,
  "setup_id": "9e5118825b2267598d5691416d3ab69e",
  "standard": "pre-standard",
  "system": {
    "driver": null,
    "hardware": "NVIDIA GB10, 128 GB unified memory",
    "os": null
  },
  "workload": {
    "cache": "cold",
    "content_tokens": 250000,
    "document": "distinct",
    "input_tokens": {
      "max": 250016,
      "median": 250016,
      "min": 250016
    },
    "output_tokens": 512,
    "repetitions": null,
    "users": 1,
    "warmup": null
  }
}