{
  "caveats": [
    "Rented single-GPU class; model and engine pins were not retained. Historical TTFT p95 withdrawn because raw samples cannot reconstruct corrected percentiles.",
    "Publisher revision was not recorded for public reproduction.",
    "Engine version was not recorded for public reproduction.",
    "Weights file SHA-256 was not recorded for public reproduction."
  ],
  "engine": {
    "name": "vLLM",
    "recipe": "plain",
    "version": null
  },
  "format": "olir/0.1",
  "history": {
    "corrections": [],
    "supersedes": []
  },
  "id": "5cae73717fc898b2755a8d10",
  "measured_at": null,
  "metrics": [
    {
      "basis": null,
      "n": null,
      "name": "requests_ok",
      "scope": "machine_total",
      "source_field": "requests_ok",
      "statistic": "single",
      "unit": "count",
      "value": 168
    },
    {
      "basis": null,
      "n": null,
      "name": "aggregate_concurrency",
      "scope": "machine_total",
      "source_field": "aggregate_concurrency",
      "statistic": "single",
      "unit": "count",
      "value": 32
    },
    {
      "basis": null,
      "n": null,
      "name": "requests_total",
      "scope": "machine_total",
      "source_field": "requests_total",
      "statistic": "single",
      "unit": "count",
      "value": 168
    },
    {
      "basis": null,
      "n": null,
      "name": "aggregate_e2e_tps",
      "scope": "machine_total",
      "source_field": "aggregate_tps",
      "statistic": "single",
      "unit": "tok/s",
      "value": 883.3
    },
    {
      "basis": null,
      "n": null,
      "name": "single_user_decode_tps",
      "scope": "per_person",
      "source_field": "single_user_decode_tps",
      "statistic": "median",
      "unit": "tok/s",
      "value": 30.7
    }
  ],
  "model": {
    "artifact_sha256": [],
    "publisher_repo": "Qwen/Qwen2.5-7B",
    "publisher_revision": null,
    "variant": "unrecorded variant",
    "weights_dtype": "unrecorded",
    "weights_format": "safetensors"
  },
  "outcome": {
    "gates": [
      {
        "met": true,
        "name": "requests",
        "observed": "168/168"
      }
    ],
    "reasons": [],
    "status": "pass"
  },
  "publication_group": "pre-standard",
  "published": null,
  "publisher": {
    "name": "Skilak Keep"
  },
  "quality": [],
  "record_date": "2026-07-16",
  "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": 8192,
    "max_output_tokens": 256
  },
  "setup_complete": false,
  "setup_id": "62a5a7d509a66a80aef56e64394d1250",
  "standard": "pre-standard",
  "system": {
    "driver": null,
    "hardware": "NVIDIA A10, 24 GB",
    "os": null
  },
  "workload": {
    "cache": null,
    "content_tokens": null,
    "document": null,
    "input_tokens": null,
    "output_tokens": 256,
    "repetitions": null,
    "users": 1,
    "warmup": null
  }
}