{
  "fresh_response_validity": {
    "cached_tokens": [
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0
    ],
    "cached_tokens_all_zero": true,
    "chat_reasoning_delta_counts": [
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0
    ],
    "classification": "fresh-response",
    "context_checkpoints_or_prefix_reuse": false,
    "each_prompt_run_once": true,
    "history_acceleration": false,
    "ngram_history_acceleration": false,
    "note": "Fixed realistic suite; each prompt is sent once as a cold response. Do not average synthetic repeated prompts, warmed continuations, or n-gram/history-accelerated rows into this metric.",
    "primary_metric_name": "median_tok_s_1_100_after_ttft",
    "primary_metric_tokens": 100,
    "prompt_count": 12,
    "prompts_are_unique": true,
    "response_reuse": false,
    "return_token_ids_requested": false,
    "suite_id": "qwen36-27b-autoround-int4-b70-realistic-v1",
    "suite_version": 1,
    "token_timing_source": "openai_stream_content_or_reasoning_delta",
    "valid": true
  },
  "output_sha256s": [
    "b2bd45c8791ba926318dbde236fe9240d315d69a50f95b3ba050131e514691d9",
    "115547429c424e9759e6aa7a034aef16e15fe398f2721622f22121f449a04a5c",
    "99821ccbf4c8518a16003e20a05555c26c244c2cca03e8e12498881120513732",
    "1fd5c17a077bbaf6490c8e15b29636f371240e06270226663e85628fca95ab30",
    "dc6a40f1d93ebe00aed38ccf75f524b88f27dfaba2203c2ebd4439a8a8099d2b",
    "53e796f0cd47a40fa4719b344d270cda9ff7809db4c7f2c857391adcf8dc659e",
    "d25294514d239e39587cb52e6791b06503f2eb8995dccbcd172ce74cf3b0baec",
    "d72a2acc1d1b28c6cd910834e3debff3e0d4d79c461b38eac2e858b3e8584251",
    "8e77401a01640e192da25efc98e5ff775b0069e242473a0534f685d263ffeb67",
    "d2a92ef0bbbefec1cc0182aec8eba3a45ce56d390f64a3d223947e82f49f40ea",
    "60b30326078bb0cd0626be0088274dcb0f91aa06418d90976f5eca55839aee11",
    "1b6bca2916680c614e32a77c393698ee91683f6b3772dd6bd207a9e9d1d32bc9"
  ],
  "prompt_sha256s": [
    "b6c5ea9e2495278fbe9314b60c34cb3eb51421c6489478812e598f7b3cc83538",
    "5a92dcc76e641515c54d71d2d3fef6e3b12a88baa616cc967eba6c897e684a33",
    "6ade325d82c3926f6a1b770d04828e63579e882b6d8de31ad31ad4a986dba43f",
    "21229aa5f10636f028248b73cc066b2d49b48580f6a005e110109cd0e4a4320b",
    "2aff657de0093a2b612f97a696a85e74cd7c7f36e1cdb1f97ae69ca44479e9e9",
    "ad01f6d7127de18807a1f11c9bfc86a2ea86fd3639c40581818dec4a7ea4664d",
    "850f7db6d007b66cb1d37440e19c25e69bf6c762c3d3a92a9b18bc3d91de1678",
    "9b645eae54d2e851f5182a77a7a68361ee52869baf8eb3cec1afe25535256c1a",
    "9fbc35260d4e350f5280a3cb75c0ae8d4371461f18086805e90dd5fb65896cb0",
    "2dcb017f7bb740ae3a8205fcb464068fcfe82189d61be31850d5e6c6492354bf",
    "07d3df497692ad49145544fc0388001879fe6a7201aca3f9eaa4e14288f0bfeb",
    "e1fcb91edf9a4e6048cd5a5738cebcc0f71edc15281607b795bf2a6dddd6adeb"
  ],
  "realistic_final_gate": {
    "cached_tokens": [
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0
    ],
    "cached_tokens_all_zero": true,
    "chunk_count_matches_completion_tokens_all": true,
    "chunk_count_matches_completion_tokens_note": "Informational only: llama.cpp usage may count an EOS/final token that is not emitted as a text delta. Promotion requires enough streamed text deltas to measure the first metric window.",
    "chunk_counts": [
      128,
      128,
      128,
      128,
      128,
      128,
      128,
      128,
      128,
      128,
      128,
      128
    ],
    "completion_tokens_at_least_metric_window": true,
    "metric_chunk_events_at_least_window": true,
    "metric_name": "median_tok_s_1_100_after_ttft",
    "metric_token_id_events_at_least_window": false,
    "metric_tokens": 100,
    "passed": true,
    "prompts_unique": true,
    "required_policy": "fixed realistic prompt suite; each prompt once; cached_tokens=0 every row; no repeated/warmed prompt averaging; metric is median tokens 1-100 after TTFT",
    "return_token_ids_requested": false,
    "stream_token_id_counts": [
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0,
      0
    ],
    "token_timing_source": "openai_stream_content_or_reasoning_delta"
  },
  "rows": [
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.343417467083782,
        1.4625344830565155,
        1.46257579093799,
        1.4625905191060156,
        1.5231337160803378,
        1.523161088116467,
        1.5231754849664867,
        1.5231910641305149,
        1.5835421411320567,
        1.5835757539607584,
        1.5835891589522362,
        1.583604358136654,
        1.64360690000467,
        1.6436328690033406,
        1.6436456330120564,
        1.7030963480938226,
        1.7031310531310737,
        1.7632864930201322,
        1.7633182620629668,
        1.7633318779990077,
        1.763356874929741,
        1.7633673751261085,
        1.823271983070299,
        1.8233048850670457,
        1.8233183601405472,
        1.8233382070902735,
        1.8233493180014193,
        1.8233664000872523,
        1.8832924589514732,
        1.8833241879474372,
        1.8833358699921519,
        1.883354616118595,
        1.8833649249281734,
        1.883382328087464,
        1.9428500449284911,
        1.9428827569354326,
        1.94289629207924,
        2.002205421915278,
        2.061734765069559,
        2.121728210011497,
        2.1217602000106126,
        2.181319158989936,
        2.1813513301312923,
        2.2407355399336666,
        2.2407674801070243,
        2.240780184045434,
        2.2407993800006807,
        2.3004874419420958,
        2.3005185399670154,
        2.3005312441382557,
        2.3005515821278095,
        2.360104559920728,
        2.360137040959671,
        2.3601512680761516,
        2.4199530130717903,
        2.4199852829333395,
        2.419998938916251,
        2.4200198180042207,
        2.420031138928607,
        2.479785986011848,
        2.4798163131345063,
        2.4798297791276127,
        2.4798496360890567,
        2.5387572119943798,
        2.5989022811409086,
        2.598932688124478,
        2.598945281933993,
        2.59896561014466,
        2.5989767010323703,
        2.658437255071476,
        2.6584813080262393,
        2.7177310860715806,
        2.717762635089457,
        2.71777488803491,
        2.777741923928261,
        2.7777753069531173,
        2.77778855105862,
        2.777809481136501,
        2.8373597140889615,
        2.837391552980989,
        2.8972416990436614,
        2.8972747910302132,
        2.8972898800857365,
        2.897310839034617,
        2.8973212391138077,
        2.956729124067351,
        3.0157282911241055,
        3.0752459331415594,
        3.135147084016353,
        3.135178894037381,
        3.1351923390757293,
        3.1943566270638257,
        3.1943963719531894,
        3.194411240052432,
        3.253394006052986,
        3.313052863115445,
        3.313085214002058,
        3.3130996010731906,
        3.372629123972729,
        3.3726613549515605,
        3.372675642138347,
        3.4321945959236473,
        3.4322261849883944,
        3.4322397499345243,
        3.4922244590707123,
        3.4922623499296606,
        3.4922759861219674,
        3.4922953620553017,
        3.551557674072683,
        3.6108801690861583,
        3.610913772135973,
        3.6109323769342154,
        3.670842626132071,
        3.670875346986577,
        3.670891728019342,
        3.6709210330154747,
        3.7305872740689665,
        3.730619333917275,
        3.7901677030604333,
        3.850284469081089,
        3.850318443030119,
        3.8503431300632656,
        3.9098130809143186,
        3.9098455319181085,
        3.966615433106199,
        3.966664785053581,
        4.0165395240765065,
        4.061009733006358
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 4.061124617932364,
      "first_text_epoch_s": 1783953502.4745817,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.7177071508485824,
      "prompt_id": "incident-retrospective",
      "prompt_index": 0,
      "prompt_sha256": "b6c5ea9e2495278fbe9314b60c34cb3eb51421c6489478812e598f7b3cc83538",
      "prompt_tokens": 87,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953505.1922874,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-00-incident-retrospective",
      "request_started_epoch_s": 1783953501.1311626,
      "response_id_first": "chatcmpl-NpEQKo3ERpEaagK2v5Rl7SAmxikuvVA9",
      "response_id_last": "chatcmpl-NpEQKo3ERpEaagK2v5Rl7SAmxikuvVA9",
      "response_x_request_id": null,
      "sha256": "b2bd45c8791ba926318dbde236fe9240d315d69a50f95b3ba050131e514691d9",
      "stream_token_id_count": 0,
      "text_preview": "**Incident Retrospective: Production API Outage Due to Cache Invalidation Failure**\n\n**Date:** October 24, 2023\n**Severity:** SEV-1\n**Duration:** 45 minutes\n**Primary Owner:** Platform Engineering Team\n\n**Executive Summary**\nOn October 24, at 14:15 UTC, our primary customer-facing API experienced a complete outage affe",
      "tok_s_1_100_after_ttft": 49.27943880864645,
      "tok_s_after_ttft_full": 47.09852566713563,
      "tok_s_wall_full": 31.51836302555239,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.343417467083782,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 87,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 215
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.199970382032916,
        1.2615331150591373,
        1.2615588831249624,
        1.2615760350599885,
        1.3214795710518956,
        1.321505580097437,
        1.3814975121058524,
        1.381522438954562,
        1.3815404630731791,
        1.441682246979326,
        1.4417072541546077,
        1.4417253080755472,
        1.441750134108588,
        1.5016478300094604,
        1.5016732071526349,
        1.5016901991330087,
        1.5612987310159951,
        1.6202343790791929,
        1.6794317190069705,
        1.6794629381038249,
        1.7389053881634027,
        1.798348949989304,
        1.79838011902757,
        1.7983939650002867,
        1.8578712900634855,
        1.91730601596646,
        1.9772445680573583,
        1.9772749750409275,
        2.0368433920666575,
        2.0965303611010313,
        2.096561640035361,
        2.096575746079907,
        2.1559516510460526,
        2.2156781351659447,
        2.275071111973375,
        2.275101870065555,
        2.334452377166599,
        2.334483755985275,
        2.3344964899588376,
        2.334515907103196,
        2.393800560152158,
        2.4532526989933103,
        2.453282575123012,
        2.5127726341597736,
        2.5128034320659935,
        2.5128173981793225,
        2.572640784084797,
        2.572674588067457,
        2.5726945651695132,
        2.631630914984271,
        2.690668805036694,
        2.7504576060455292,
        2.7504888139665127,
        2.7505024201236665,
        2.8100190691184253,
        2.8100500979926437,
        2.810064635006711,
        2.869127321988344,
        2.9281837071757764,
        2.9877362330444157,
        2.98776714107953,
        2.987779224058613,
        3.0474815231282264,
        3.047512110089883,
        3.1068490201141685,
        3.16662465617992,
        3.166656055022031,
        3.166669380152598,
        3.225783152040094,
        3.2258132779970765,
        3.2848329241387546,
        3.3439048670697957,
        3.403521885164082,
        3.4035523820202798,
        3.463751824107021,
        3.463784154970199,
        3.463798141106963,
        3.4638175980653614,
        3.463835101108998,
        3.4638724110554904,
        3.523577944142744,
        3.5833968019578606,
        3.5834279300179332,
        3.583441555965692,
        3.6429159760009497,
        3.6429467941634357,
        3.702835442032665,
        3.7028667910490185,
        3.7028801161795855,
        3.702899171039462,
        3.7623574209865183,
        3.7623887700028718,
        3.821365586016327,
        3.881632364122197,
        3.8816641240846366,
        3.881678050151095,
        3.881698067067191,
        3.881715781055391,
        3.8817494539543986,
        3.9418406530749053,
        3.9418730239849538,
        3.94189569610171,
        3.9419132489711046,
        3.9419308120850474,
        4.00149777601473,
        4.001527582062408,
        4.061124402098358,
        4.061155290110037,
        4.121321058133617,
        4.121353558963165,
        4.121365792118013,
        4.121385590173304,
        4.121404365170747,
        4.121437167050317,
        4.181344149168581,
        4.18137473706156,
        4.181388893164694,
        4.181410133140162,
        4.240731545956805,
        4.2407520650886,
        4.300176321063191,
        4.300218851072714,
        4.300242054043338,
        4.300261900993064,
        4.357482017017901,
        4.357502676080912,
        4.357515320181847,
        4.357526651117951
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 4.357614926993847,
      "first_text_epoch_s": 1783953506.3923736,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 3.157644544960931,
      "prompt_id": "code-review",
      "prompt_index": 1,
      "prompt_sha256": "5a92dcc76e641515c54d71d2d3fef6e3b12a88baa616cc967eba6c897e684a33",
      "prompt_tokens": 90,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953509.5500154,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-01-code-review",
      "request_started_epoch_s": 1783953505.192401,
      "response_id_first": "chatcmpl-xtDMGGqDNzhgYaBYO6vmq6CRzjGDExuY",
      "response_id_last": "chatcmpl-xtDMGGqDNzhgYaBYO6vmq6CRzjGDExuY",
      "response_x_request_id": null,
      "sha256": "115547429c424e9759e6aa7a034aef16e15fe398f2721622f22121f449a04a5c",
      "stream_token_id_count": 0,
      "text_preview": "This is a classic \"performance vs. reliability\" trade-off pattern. While batching improves throughput, the specific implementation details you described introduce significant risks regarding data integrity and operational visibility.\n\nBelow is a detailed code review structured for a Pull Request (PR).\n\n### **Verdict: R",
      "tok_s_1_100_after_ttft": 36.47145565424477,
      "tok_s_after_ttft_full": 40.536544939571,
      "tok_s_wall_full": 29.373866701044726,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.199970382032916,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 90,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 218
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.151226415997371,
        1.2127258700784296,
        1.2127541829831898,
        1.2127679591067135,
        1.212782897055149,
        1.2724821390584111,
        1.3322576449718326,
        1.3322831930126995,
        1.3322965679690242,
        1.3916713709477335,
        1.4512234469875693,
        1.4512493261136115,
        1.4512630719691515,
        1.451278069987893,
        1.5110482049640268,
        1.5110842131543905,
        1.5110982190817595,
        1.5709078391082585,
        1.5709424640517682,
        1.5709560089744627,
        1.5709768990054727,
        1.570988150080666,
        1.6310168709605932,
        1.6310530989430845,
        1.6310660340823233,
        1.6310855001211166,
        1.6310951779596508,
        1.6906719710677862,
        1.690702788066119,
        1.7502337850164622,
        1.7502649340312928,
        1.8106704021338373,
        1.8106898490805179,
        1.8107160180807114,
        1.8107278801035136,
        1.8107377090491354,
        1.8107550111599267,
        1.86993601010181,
        1.8699677200056612,
        1.9292495481204242,
        1.9892596150748432,
        1.989294660044834,
        1.9893081160262227,
        1.9893181750085205,
        1.9893360279966146,
        1.9893455361016095,
        2.0486869460437447,
        2.108178398106247,
        2.108209507074207,
        2.1678951440844685,
        2.1679277550429106,
        2.167942252010107,
        2.2282394180074334,
        2.2282713579479605,
        2.2282848830800503,
        2.2283032471314073,
        2.2283139880746603,
        2.228332262020558,
        2.287632034160197,
        2.2876619501039386,
        2.2876752950251102,
        2.287695433013141,
        2.3467001309618354,
        2.4068063881713897,
        2.4068266360554844,
        2.4068411129992455,
        2.4068532960955054,
        2.465995952952653,
        2.525851228972897,
        2.5258865549694747,
        2.5259027250576764,
        2.5259140070993453,
        2.5259235850535333,
        2.5849810920190066,
        2.5850124610587955,
        2.585026988061145,
        2.644020585110411,
        2.7038448320236057,
        2.7038746480830014,
        2.7038875319994986,
        2.703907269053161,
        2.7039167070761323,
        2.7638742551207542,
        2.7639056839980185,
        2.7639178871177137,
        2.763935670023784,
        2.7639466810505837,
        2.823852820089087,
        2.8238855821546167,
        2.8238992169499397,
        2.823917993111536,
        2.823928182013333,
        2.8837806889787316,
        2.883814202155918,
        2.8838283880613744,
        2.883847394026816,
        2.8838567819911987,
        2.9436156630981714,
        2.943647873122245,
        2.9436613279394805,
        2.9436793620698154,
        3.0028128779958934,
        3.002843827009201,
        3.0622150890994817,
        3.062246478162706,
        3.1213938400615007,
        3.1214239271357656,
        3.181105172028765,
        3.1811364199966192,
        3.1811497861053795,
        3.2400891871657223,
        3.2999597180169076,
        3.2999913671519607,
        3.3000046720262617,
        3.3000237981323153,
        3.3593044499866664,
        3.418876781128347,
        3.418906917097047,
        3.4779724089894444,
        3.4780023160856217,
        3.537675205152482,
        3.537694290978834,
        3.537727212999016,
        3.6008269730955362,
        3.6008549351245165,
        3.6008696029894054,
        3.6008806941099465,
        3.600891655078158
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 3.6009697620756924,
      "first_text_epoch_s": 1783953510.7013407,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.4497433460783213,
      "prompt_id": "customer-email",
      "prompt_index": 2,
      "prompt_sha256": "6ade325d82c3926f6a1b770d04828e63579e882b6d8de31ad31ad4a986dba43f",
      "prompt_tokens": 77,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953513.1510825,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-02-customer-email",
      "request_started_epoch_s": 1783953509.5501127,
      "response_id_first": "chatcmpl-JcLlwlsM5uLLiMhN9s5ix4gWi5Ahl3v4",
      "response_id_last": "chatcmpl-JcLlwlsM5uLLiMhN9s5ix4gWi5Ahl3v4",
      "response_x_request_id": null,
      "sha256": "99821ccbf4c8518a16003e20a05555c26c244c2cca03e8e12498881120513732",
      "stream_token_id_count": 0,
      "text_preview": "Subject: Update on the status of the [Feature Name] export feature\n\nDear [Customer Name],\n\nI am writing to provide an update on the release of the [Feature Name] export functionality, which was originally scheduled for [Original Date].\n\nDuring our final validation phase, our engineering team identified a data-integrity",
      "tok_s_1_100_after_ttft": 55.790031389005726,
      "tok_s_after_ttft_full": 52.25037153582198,
      "tok_s_wall_full": 35.545980237894995,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.151226415997371,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 77,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 205
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.15580328903161,
        1.2169559069443494,
        1.2766935769468546,
        1.2767196360509843,
        1.336731222923845,
        1.3367571020498872,
        1.3367704469710588,
        1.3367852149531245,
        1.3367978490423411,
        1.3966588210314512,
        1.3966840389184654,
        1.456694063032046,
        1.4567191998939961,
        1.5162962379399687,
        1.5752717410214245,
        1.6350914570502937,
        1.6351277648936957,
        1.6351419719867408,
        1.6946414830163121,
        1.6946724718436599,
        1.7544592248741537,
        1.7544894618913531,
        1.754502696916461,
        1.7545250689145178,
        1.7545374219771475,
        1.7545563578605652,
        1.8134795329533517,
        1.8735995038878173,
        1.8736306428909302,
        1.873645189916715,
        1.9325792149174958,
        1.9922875589691103,
        1.992318085860461,
        2.052189178066328,
        2.0522223298903555,
        2.0522360058967024,
        2.1120630940422416,
        2.112092088907957,
        2.1121044519823045,
        2.1121242300141603,
        2.171673353994265,
        2.171706325840205,
        2.1717197210527956,
        2.2313847239129245,
        2.23141654394567,
        2.2314410598482937,
        2.2908082520589232,
        2.2908412739634514,
        2.2908543690573424,
        2.3502785868477076,
        2.3503094860352576,
        2.4101695858407766,
        2.4102019870188087,
        2.4102162539493293,
        2.4102348790038377,
        2.4698801040649414,
        2.469911092892289,
        2.528766739880666,
        2.588200486963615,
        2.588232706999406,
        2.5882456719409674,
        2.647845732048154,
        2.647885357029736,
        2.707207994069904,
        2.767165988916531,
        2.7671849648468196,
        2.7671995719429106,
        2.7672111140564084,
        2.767222546041012,
        2.8261793728452176,
        2.885337610030547,
        2.885369700845331,
        2.9451722018420696,
        2.9452041620388627,
        2.9452176080085337,
        2.9452393089886755,
        2.945250720018521,
        3.0049700250383466,
        3.005002294899896,
        3.0050148088485003,
        3.005035076988861,
        3.005045606987551,
        3.0649488079361618,
        3.0649803068954498,
        3.0649923000019044,
        3.0650118270423263,
        3.065022567054257,
        3.06504086102359,
        3.1250173510052264,
        3.1250494320411235,
        3.1250629969872534,
        3.1250842069275677,
        3.1250954680144787,
        3.1251146639697254,
        3.1842394778504968,
        3.184271087870002,
        3.2441192748956382,
        3.244150514015928,
        3.2441645900253206,
        3.2441848379094154,
        3.2441961399745196,
        3.304086216026917,
        3.3041185268666595,
        3.3041316620074213,
        3.3041503969579935,
        3.3637851518578827,
        3.3638182538561523,
        3.3638326718937606,
        3.4238626819569618,
        3.4238950428552926,
        3.423908457858488,
        3.4239297779276967,
        3.4239421610254794,
        3.4239597250707448,
        3.482793188886717,
        3.5421616318635643,
        3.5421927410643548,
        3.542207788908854,
        3.6013250078540295,
        3.6013569380156696,
        3.6611239130143076,
        3.6611566350329667,
        3.6611706810072064,
        3.661190317943692,
        3.661200497066602,
        3.7147250128909945,
        3.714744418859482,
        3.714758755872026
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 3.714844848960638,
      "first_text_epoch_s": 1783953514.306981,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.559041559929028,
      "prompt_id": "sql-debugging",
      "prompt_index": 3,
      "prompt_sha256": "21229aa5f10636f028248b73cc066b2d49b48580f6a005e110109cd0e4a4320b",
      "prompt_tokens": 83,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953516.8660214,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-03-sql-debugging",
      "request_started_epoch_s": 1783953513.1511767,
      "response_id_first": "chatcmpl-6N5y93ohmhCyPqb7yX57C6PXsJ7decKk",
      "response_id_last": "chatcmpl-6N5y93ohmhCyPqb7yX57C6PXsJ7decKk",
      "response_x_request_id": null,
      "sha256": "1fd5c17a077bbaf6490c8e15b29636f371240e06270226663e85628fca95ab30",
      "stream_token_id_count": 0,
      "text_preview": "Here is a comprehensive debugging plan for diagnosing and resolving the performance regression in your PostgreSQL reporting query following a schema migration.\n\n### 1. Immediate Diagnosis: `EXPLAIN (ANALYZE, BUFFERS, FORMAT TEXT)`\n\nStart by capturing the actual execution plan with real-world metrics. Do not rely on `EX",
      "tok_s_1_100_after_ttft": 47.883970270535634,
      "tok_s_after_ttft_full": 50.01872654367908,
      "tok_s_wall_full": 34.45635153129279,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.15580328903161,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 83,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 211
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.1541112679988146,
        1.2152389958500862,
        1.2753019868396223,
        1.2753278559539467,
        1.3354670300614089,
        1.3355059940367937,
        1.3355209920555353,
        1.395515674026683,
        1.3955417729448527,
        1.3955556389410049,
        1.4551991689950228,
        1.514701253036037,
        1.5743435609620064,
        1.634786827955395,
        1.6348139790352434,
        1.6348274250049144,
        1.6348518908489496,
        1.6348703559488058,
        1.6348879588767886,
        1.6942614899016917,
        1.6942939418368042,
        1.6943073968868703,
        1.7547201870474964,
        1.7547394230496138,
        1.7547524380497634,
        1.754763217875734,
        1.7547725460026413,
        1.7547811118420213,
        1.8143018600530922,
        1.814333108952269,
        1.8143473458476365,
        1.8737018208485097,
        1.8737327388953418,
        1.9330411578994244,
        1.9330729879438877,
        1.9330866439267993,
        1.9922424340620637,
        1.9922737630549818,
        2.0517960749566555,
        2.0518281350377947,
        2.110837339889258,
        2.169985083863139,
        2.229711340041831,
        2.2297413658816367,
        2.2297540300060064,
        2.229774748906493,
        2.289215025026351,
        2.3491359769832343,
        2.3491682279855013,
        2.3491815528832376,
        2.408404520014301,
        2.467731051845476,
        2.4677716880105436,
        2.4677848629653454,
        2.5272637819871306,
        2.5272946408949792,
        2.5865396689623594,
        2.586575075984001,
        2.646655537886545,
        2.6466856149490923,
        2.6466979479882866,
        2.6467197488527745,
        2.646731582004577,
        2.6467508680652827,
        2.7061533629894257,
        2.706185664050281,
        2.7654295100364834,
        2.7654617810621858,
        2.765474824933335,
        2.8250554059632123,
        2.825095140840858,
        2.8251135060563684,
        2.884856892982498,
        2.8848760589025915,
        2.8849012970458716,
        2.884913839865476,
        2.9441541398409754,
        3.0035360460169613,
        3.0035677049309015,
        3.003581781871617,
        3.0631992819253355,
        3.063231793930754,
        3.0632447879761457,
        3.1223903989885002,
        3.1224211868830025,
        3.181537572061643,
        3.2410746898967773,
        3.300713158911094,
        3.300744557986036,
        3.3007565909065306,
        3.359731188043952,
        3.4192951768636703,
        3.4193267568480223,
        3.4787049549631774,
        3.478735974058509,
        3.5387634958606213,
        3.5387831630650908,
        3.538796267937869,
        3.597929724957794,
        3.5979608439374715,
        3.657674974994734,
        3.7171753540169448,
        3.7774041059892625,
        3.7774366070516407,
        3.777450372930616,
        3.7774704908952117,
        3.7774806700181216,
        3.777514975052327,
        3.836810997920111,
        3.8968953678850085,
        3.896928030066192,
        3.8969401528593153,
        3.9564232488628477,
        3.9564616109710187,
        4.016095933038741,
        4.016128242947161,
        4.0161419690120965,
        4.076077577890828,
        4.076108696870506,
        4.076122011989355,
        4.076142451027408,
        4.076153340982273,
        4.135467639891431,
        4.197377539938316,
        4.197398578980938,
        4.197413207031786,
        4.247818027855828,
        4.247845089063048
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 4.247937332838774,
      "first_text_epoch_s": 1783953518.0202248,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 3.093826064839959,
      "prompt_id": "release-plan",
      "prompt_index": 4,
      "prompt_sha256": "2aff657de0093a2b612f97a696a85e74cd7c7f36e1cdb1f97ae69ca44479e9e9",
      "prompt_tokens": 68,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953521.1140494,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-04-release-plan",
      "request_started_epoch_s": 1783953516.8661122,
      "response_id_first": "chatcmpl-wtKD98jSDYc9sPr5PK2rz97W1thyJQaj",
      "response_id_last": "chatcmpl-wtKD98jSDYc9sPr5PK2rz97W1thyJQaj",
      "response_x_request_id": null,
      "sha256": "dc6a40f1d93ebe00aed38ccf75f524b88f27dfaba2203c2ebd4439a8a8099d2b",
      "stream_token_id_count": 0,
      "text_preview": "# Engineering Release Plan: New Authentication Flow (Project: AuthNext)\n\n**Date:** October 26, 2023\n**Owner:** Platform Engineering Team\n**Status:** Draft / For Review\n**Target Release Date:** Q4 2023\n\n## 1. Executive Summary\nThis document outlines the technical strategy for deploying the new authentication flow (\"Auth",
      "tok_s_1_100_after_ttft": 40.919048776392486,
      "tok_s_after_ttft_full": 41.372720158597964,
      "tok_s_wall_full": 30.13227125798047,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.1541112679988146,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 68,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 196
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.1519269400741905,
        1.2134323369245976,
        1.2134590479545295,
        1.213486498920247,
        1.213502359110862,
        1.2734379980247468,
        1.273471751017496,
        1.3336585420183837,
        1.3336843510624021,
        1.3336979569867253,
        1.3337123841047287,
        1.3337260000407696,
        1.393498970894143,
        1.3935652859508991,
        1.4530258679296821,
        1.5126409919466823,
        1.5126752669457346,
        1.57247577002272,
        1.5725092228967696,
        1.5725224879570305,
        1.5725435470230877,
        1.63210494001396,
        1.6321434821002185,
        1.6911466119345278,
        1.7512580519542098,
        1.7512881578877568,
        1.7513024848885834,
        1.7513239660765976,
        1.8106150990352035,
        1.869673252105713,
        1.8697038399986923,
        1.9294882121030241,
        1.9295211140997708,
        1.9295355009380728,
        1.9889569289516658,
        2.049130063969642,
        2.109255458926782,
        2.1691951940301806,
        2.229061079910025,
        2.22908080695197,
        2.229092950001359,
        2.2883895121049136,
        2.348210483090952,
        2.4079726829659194,
        2.408005554927513,
        2.408019321039319,
        2.4683500931132585,
        2.46838232409209,
        2.468396390089765,
        2.4684161480981857,
        2.468426486942917,
        2.46844388009049,
        2.5277267969213426,
        2.5277591480407864,
        2.5277740960009396,
        2.5872029380407184,
        2.587233225116506,
        2.6467904578894377,
        2.6468222478870302,
        2.706747565884143,
        2.70677994703874,
        2.706794124096632,
        2.706814863020554,
        2.7068247620481998,
        2.706841523060575,
        2.7662079269066453,
        2.826856479048729,
        2.8268764060921967,
        2.8268898820970207,
        2.8269007420167327,
        2.826911702984944,
        2.8269219419453293,
        2.886951986933127,
        2.8869823440909386,
        2.886996360961348,
        2.8870177909266204,
        2.887028300901875,
        2.9458707969170064,
        3.005905339960009,
        3.0059245659504086,
        3.005948852049187,
        3.005960243055597,
        3.065255692927167,
        3.0652876729145646,
        3.0653031019028276,
        3.12468283995986,
        3.184849132085219,
        3.1848821840249,
        3.1848962609656155,
        3.184916378930211,
        3.184926758054644,
        3.1849434298928827,
        3.2441088939085603,
        3.303402880905196,
        3.3034334890544415,
        3.363201508997008,
        3.3632326680235565,
        3.363246433902532,
        3.4234056419227272,
        3.4234370009507984,
        3.4234506071079522,
        3.4234815349336714,
        3.423492836067453,
        3.4235113610047847,
        3.482641078066081,
        3.482672557933256,
        3.5421276479028165,
        3.601452693110332,
        3.6613822490908206,
        3.6614138178993016,
        3.661427513929084,
        3.661447871942073,
        3.7209092450793833,
        3.7806683878879994,
        3.780699155991897,
        3.7807137239724398,
        3.8399852078873664,
        3.8400168479420245,
        3.9000272350385785,
        3.9000591449439526,
        3.9000730619300157,
        3.900093099102378,
        3.9001042200252414,
        3.900121121900156,
        3.956764582078904,
        3.9567975939717144,
        3.956812391988933,
        3.9983144679572433
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 3.998406020924449,
      "first_text_epoch_s": 1783953522.2660713,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.8464790808502585,
      "prompt_id": "benchmark-analysis",
      "prompt_index": 5,
      "prompt_sha256": "ad01f6d7127de18807a1f11c9bfc86a2ea86fd3639c40581818dec4a7ea4664d",
      "prompt_tokens": 65,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953525.112549,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-05-benchmark-analysis",
      "request_started_epoch_s": 1783953521.1141431,
      "response_id_first": "chatcmpl-oOy9uGarwivdzPW7kxWG9QhStG88WHPG",
      "response_id_last": "chatcmpl-oOy9uGarwivdzPW7kxWG9QhStG88WHPG",
      "response_x_request_id": null,
      "sha256": "53e796f0cd47a40fa4719b344d270cda9ff7809db4c7f2c857391adcf8dc659e",
      "stream_token_id_count": 0,
      "text_preview": "This is a classic scenario in Large Language Model (LLM) serving benchmarks, often referred to as the **\"Prefill vs. Decode\"** performance gap, exacerbated by **KV Cache warming** and **GPU memory caching**.\n\nBelow is a detailed analysis of the causes, methodologies to isolate true throughput, and best practices for re",
      "tok_s_1_100_after_ttft": 44.02357784909329,
      "tok_s_after_ttft_full": 44.967834424332295,
      "tok_s_wall_full": 32.01275691616877,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.1519269400741905,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 65,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 193
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.1572188308928162,
        1.218466870021075,
        1.218493599910289,
        1.2185083979275078,
        1.2782415519468486,
        1.3381364298984408,
        1.33816231880337,
        1.3381759847979993,
        1.3976643360219896,
        1.3976900749839842,
        1.4580656089819968,
        1.4580921488814056,
        1.4581057149916887,
        1.4581200319807976,
        1.4581335969269276,
        1.4581566308625042,
        1.5180841498076916,
        1.5181166918482631,
        1.518130627926439,
        1.5775487879291177,
        1.5775885628536344,
        1.6379486278165132,
        1.6379927308298647,
        1.638007378904149,
        1.6380261739250273,
        1.6380371048580855,
        1.6380542169790715,
        1.6979197098407894,
        1.6979633518494666,
        1.697978189913556,
        1.6980072748847306,
        1.7575190810021013,
        1.7575660289730877,
        1.81713568395935,
        1.8171675838530064,
        1.8171823918819427,
        1.876596714835614,
        1.9362570298835635,
        1.936289079952985,
        1.9363028758671135,
        1.9955435108859092,
        1.995575290871784,
        2.0552545718383044,
        2.055286781862378,
        2.05530130979605,
        2.055321777937934,
        2.11479783593677,
        2.1744236359372735,
        2.174456408014521,
        2.174470794852823,
        2.234297564951703,
        2.2343299959320575,
        2.2343448838219047,
        2.234365101903677,
        2.294089367846027,
        2.2941215080209076,
        2.294135524891317,
        2.2941568149253726,
        2.2941669239662588,
        2.3535198499448597,
        2.353552351007238,
        2.4132323029916734,
        2.413265174953267,
        2.4132801529485732,
        2.473240864928812,
        2.4732742479536682,
        2.4732876929920167,
        2.5331247018184513,
        2.533156472025439,
        2.5331689848098904,
        2.533189092995599,
        2.5925760730169713,
        2.5926077829208225,
        2.651928056962788,
        2.7121934129390866,
        2.712226314935833,
        2.712241042871028,
        2.7122606199700385,
        2.712271299911663,
        2.712288541952148,
        2.772393314866349,
        2.772425014991313,
        2.772439011838287,
        2.772459569852799,
        2.7724710619077086,
        2.8322894948069006,
        2.8323191709350795,
        2.832332415971905,
        2.8323523339349777,
        2.891879788832739,
        2.891910646809265,
        2.9518370230216533,
        2.9518696239683777,
        2.9518842219840735,
        2.951903288019821,
        3.011895988835022,
        3.011928299907595,
        3.0119414150249213,
        3.0119796069338918,
        3.011991980019957,
        3.0714700208045542,
        3.071501289959997,
        3.1311435708776116,
        3.190681994892657,
        3.190713534830138,
        3.2507110747974366,
        3.2507439767941833,
        3.250757752917707,
        3.250778201967478,
        3.310883594909683,
        3.3109166878275573,
        3.3109305140096694,
        3.3109640169423074,
        3.3109757588244975,
        3.3705702798906714,
        3.370602269889787,
        3.3706171279773116,
        3.4302641269750893,
        3.4302836938295513,
        3.489870900986716,
        3.4899028309155256,
        3.5495562029536813,
        3.549588192952797,
        3.6115680600050837,
        3.611603997880593,
        3.6116223528515548,
        3.6116348658688366,
        3.6531455868389457
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 3.653235907899216,
      "first_text_epoch_s": 1783953526.2698753,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.4960170770063996,
      "prompt_id": "architecture-tradeoff",
      "prompt_index": 6,
      "prompt_sha256": "850f7db6d007b66cb1d37440e19c25e69bf6c762c3d3a92a9b18bc3d91de1678",
      "prompt_tokens": 67,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953528.765891,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-06-architecture-tradeoff",
      "request_started_epoch_s": 1783953525.1126552,
      "response_id_first": "chatcmpl-m1QfDO6NdXN6FCH1Agb5yG4QtZD9nBu7",
      "response_id_last": "chatcmpl-m1QfDO6NdXN6FCH1Agb5yG4QtZD9nBu7",
      "response_x_request_id": null,
      "sha256": "d25294514d239e39587cb52e6791b06503f2eb8995dccbcd172ce74cf3b0baec",
      "stream_token_id_count": 0,
      "text_preview": "This is a classic systems engineering trade-off in Large Language Model (LLM) serving. The choice between **Tensor Parallelism (TP)** and **Replica Parallelism (Independent GPUs)** depends heavily on the model size, hardware topology, and workload characteristics.\n\nBelow is a detailed comparison across the requested di",
      "tok_s_1_100_after_ttft": 53.91494913923041,
      "tok_s_after_ttft_full": 51.28170042551027,
      "tok_s_wall_full": 35.037430712654434,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.1572188308928162,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 67,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 195
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.153397458139807,
        1.2153227010276169,
        1.2153495920356363,
        1.2153642200864851,
        1.2153796690981835,
        1.2750919500831515,
        1.334815511945635,
        1.3348352699540555,
        1.393982025096193,
        1.4534581010229886,
        1.5130751221440732,
        1.5727228119503707,
        1.572749081067741,
        1.5727630171459168,
        1.6321423810441047,
        1.6321792700327933,
        1.6321933269500732,
        1.6917781981173903,
        1.691810259129852,
        1.6918241251260042,
        1.751547907013446,
        1.751580768963322,
        1.7515946559142321,
        1.7516152339521796,
        1.8108704339247197,
        1.870550444116816,
        1.929784633917734,
        1.9895889989566058,
        2.049528158036992,
        2.0495594469830394,
        2.049572350922972,
        2.049592409050092,
        2.1090970791410655,
        2.109128588112071,
        2.1688679500948638,
        2.1689001710619777,
        2.1689141469541937,
        2.2287237921264023,
        2.2287548009771854,
        2.2881334719713777,
        2.348015703028068,
        2.3480489561334252,
        2.3480625411029905,
        2.4075390370562673,
        2.4675141230691224,
        2.4675468450877815,
        2.467560169985518,
        2.527324428083375,
        2.527356087928638,
        2.527369884075597,
        2.527390713104978,
        2.5875352879520506,
        2.5875676090363413,
        2.587581606116146,
        2.6471269009634852,
        2.6471591519657522,
        2.7065383549779654,
        2.765968322986737,
        2.8258698999416083,
        2.825902973068878,
        2.8259166991338134,
        2.8858125149272382,
        2.945987347047776,
        2.946020821109414,
        2.9460353481117636,
        2.946056307060644,
        2.9460668270476162,
        2.9460832779295743,
        3.005629836115986,
        3.0056492229923606,
        3.065406318055466,
        3.0654393299482763,
        3.06545362691395,
        3.125083482125774,
        3.1251143401023,
        3.185096269007772,
        3.1851284690201283,
        3.185142496135086,
        3.185161310946569,
        3.2444994561374187,
        3.3043547261040658,
        3.304385453928262,
        3.3043988689314574,
        3.364118673140183,
        3.364150332985446,
        3.36416266602464,
        3.423959464998916,
        3.423990914132446,
        3.424005551962182,
        3.4837669839616865,
        3.483804624993354,
        3.543603368103504,
        3.5436523801181465,
        3.5436737309210002,
        3.543739103944972,
        3.60383483604528,
        3.603865975048393,
        3.6038808829616755,
        3.6039003799669445,
        3.6039115709718317,
        3.603928613010794,
        3.6637777311261743,
        3.663809149991721,
        3.6638239880558103,
        3.663843885064125,
        3.7239500279538333,
        3.723981938092038,
        3.723996073938906,
        3.7240153709426522,
        3.724026191048324,
        3.783577447058633,
        3.7836091071367264,
        3.8438190550077707,
        3.8438416570425034,
        3.8438567460980266,
        3.843868477968499,
        3.8438780361320823,
        3.9039310179650784,
        3.903950445121154,
        3.9039646110031754,
        3.9039763731416315,
        3.903986281948164,
        3.9636501311324537,
        3.963681360008195,
        3.9636945449747145,
        3.9637128990143538,
        4.014111963100731,
        4.014143072068691
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 4.014237368945032,
      "first_text_epoch_s": 1783953529.9193866,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.8608399108052254,
      "prompt_id": "bug-report-synthesis",
      "prompt_index": 7,
      "prompt_sha256": "9b645eae54d2e851f5182a77a7a68361ee52869baf8eb3cec1afe25535256c1a",
      "prompt_tokens": 71,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953532.780225,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-07-bug-report-synthesis",
      "request_started_epoch_s": 1783953528.7659879,
      "response_id_first": "chatcmpl-NKB4XgxLHJwTniWu0i0unWytRQZIDImY",
      "response_id_last": "chatcmpl-NKB4XgxLHJwTniWu0i0unWytRQZIDImY",
      "response_x_request_id": null,
      "sha256": "d72a2acc1d1b28c6cd910834e3debff3e0d4d79c461b38eac2e858b3e8584251",
      "stream_token_id_count": 0,
      "text_preview": "# Bug Report: Race Condition in Token Cache Under Load\n\n## Summary\nUsers occasionally receive responses intended for previous requests. This issue manifests exclusively under high load conditions. Server logs indicate cache hits for requests that should have `cached_tokens = 0`, suggesting a cache coherence or race con",
      "tok_s_1_100_after_ttft": 40.80776334906776,
      "tok_s_after_ttft_full": 44.74210511275079,
      "tok_s_wall_full": 31.886505015930144,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.153397458139807,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 71,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 199
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.1526216708589345,
        1.2144753569737077,
        1.2145259820390493,
        1.2145643238909543,
        1.2145959038753062,
        1.2146223939489573,
        1.2146466800477356,
        1.2743752079550177,
        1.2744236099533737,
        1.2744565319735557,
        1.3338646469637752,
        1.333917706971988,
        1.3931731539778411,
        1.3932100129313767,
        1.4529157800134271,
        1.4529648320749402,
        1.5125336800701916,
        1.512593121966347,
        1.512619521934539,
        1.5717705618590117,
        1.5718291129451245,
        1.6308133490383625,
        1.6911567670758814,
        1.691218402935192,
        1.6912452238611877,
        1.691306479042396,
        1.6913286410272121,
        1.6914007070008665,
        1.7506385499145836,
        1.7506978120654821,
        1.7507242618594319,
        1.810363191878423,
        1.810428065015003,
        1.8104544940870255,
        1.8104864149354398,
        1.8699998480733484,
        1.870064128888771,
        1.870092883007601,
        1.9299252980854362,
        1.9299845099449158,
        1.9300122219137847,
        1.9893423998728395,
        1.9893881159368902,
        2.0490814789664,
        2.0491417918819934,
        2.049169555073604,
        2.1085268340539187,
        2.1680064238607883,
        2.227900985861197,
        2.2279614398721606,
        2.227986726909876,
        2.2280213618651032,
        2.287793722935021,
        2.287853655871004,
        2.28788248100318,
        2.2879231569822878,
        2.287943755974993,
        2.347229529870674,
        2.406619920860976,
        2.4662467068992555,
        2.4663054479751736,
        2.5259253510739654,
        2.5259832609444857,
        2.5854071548674256,
        2.6452266939450055,
        2.6452664490789175,
        2.7051530960015953,
        2.7051881218794733,
        2.7052028400357813,
        2.7052160538733006,
        2.7646467408630997,
        2.8246625009924173,
        2.8246927680447698,
        2.8247056019026786,
        2.824725759914145,
        2.885530195897445,
        2.8855706630274653,
        2.885589297860861,
        2.9466162140015513,
        2.9466536240652204,
        2.9466757860500365,
        2.9466945610474795,
        3.0067924060858786,
        3.0068221520632505,
        3.006834795931354,
        3.0068539220374078,
        3.0068720360286534,
        3.0670170290395617,
        3.067048868862912,
        3.067061211913824,
        3.067080267937854,
        3.067092281067744,
        3.126682589063421,
        3.1267147690523416,
        3.186630350071937,
        3.186663281871006,
        3.186677108053118,
        3.2466555880382657,
        3.246687988983467,
        3.2467017450835556,
        3.246721181087196,
        3.3064331479836255,
        3.306464276043698,
        3.365950306877494,
        3.42578499391675,
        3.425816534087062,
        3.4258300489746034,
        3.485258861910552,
        3.5455682040192187,
        3.545589284040034,
        3.54561648494564,
        3.5456357109360397,
        3.6056906040757895,
        3.6057224238757044,
        3.605736450990662,
        3.6057562679052353,
        3.6663049119524658,
        3.666337131988257,
        3.6663492049556226,
        3.6663690330460668,
        3.666380665032193,
        3.6663985878694803,
        3.7258265090640634,
        3.725865873042494,
        3.7830311390571296,
        3.783053651917726,
        3.783067046897486,
        3.783078007865697
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 3.7831908599473536,
      "first_text_epoch_s": 1783953533.9329479,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.630569189088419,
      "prompt_id": "technical-guide",
      "prompt_index": 8,
      "prompt_sha256": "9fbc35260d4e350f5280a3cb75c0ae8d4371461f18086805e90dd5fb65896cb0",
      "prompt_tokens": 69,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953536.5635152,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-08-technical-guide",
      "request_started_epoch_s": 1783953532.7803245,
      "response_id_first": "chatcmpl-e5yBt1IIfZoyD0xYO9y0dher9oM4YYxc",
      "response_id_last": "chatcmpl-e5yBt1IIfZoyD0xYO9y0dher9oM4YYxc",
      "response_x_request_id": null,
      "sha256": "8e77401a01640e192da25efc98e5ff775b0069e242473a0534f685d263ffeb67",
      "stream_token_id_count": 0,
      "text_preview": "# Technical Guide: Adding a New Benchmark to the Optimization Repository\n\nThis guide outlines the standard procedure for integrating a new benchmark into the existing optimization repository. Adhering to these conventions ensures reproducibility, ease of maintenance, and consistent result aggregation.\n\n## 1. Directory ",
      "tok_s_1_100_after_ttft": 47.753665789034926,
      "tok_s_after_ttft_full": 48.6586707283515,
      "tok_s_wall_full": 33.83387324047966,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.1526216708589345,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 69,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 197
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.1648355519864708,
        1.226779795018956,
        1.2268440059851855,
        1.2268806658685207,
        1.226902776863426,
        1.226929206866771,
        1.2269545949529856,
        1.2865502110216767,
        1.346732131903991,
        1.3467818559147418,
        1.3468066630885005,
        1.3468370898626745,
        1.4069065400399268,
        1.406959149055183,
        1.4664866970852017,
        1.4665586028713733,
        1.4665934778749943,
        1.5259232418611646,
        1.5259958989918232,
        1.585590282920748,
        1.585646518971771,
        1.6455118330195546,
        1.6455714248586446,
        1.645598747069016,
        1.6456459760665894,
        1.704797884915024,
        1.7048553130589426,
        1.764047228032723,
        1.8235860159620643,
        1.8236445670481771,
        1.82367756892927,
        1.8831456140615046,
        1.8832022310234606,
        1.9428327530622482,
        1.9428908019326627,
        1.9429231330286711,
        2.0027001700364053,
        2.002759332070127,
        2.002786102006212,
        2.0028373289387673,
        2.0623488558921963,
        2.0623922580853105,
        2.1217451859265566,
        2.1813024289440364,
        2.181359878042713,
        2.2412291490472853,
        2.241298499982804,
        2.241326994029805,
        2.241347522009164,
        2.300276542082429,
        2.300333749037236,
        2.3602165458723903,
        2.3602753370068967,
        2.3603011360391974,
        2.3603520719334483,
        2.419712683884427,
        2.4197722170501947,
        2.4198051490820944,
        2.4790023430250585,
        2.538814866915345,
        2.538873466895893,
        2.5388994459062815,
        2.538935183081776,
        2.598527983063832,
        2.59858693392016,
        2.598625225946307,
        2.658305700868368,
        2.6583649930544198,
        2.658397044055164,
        2.7182683679275215,
        2.718327600043267,
        2.7183637879788876,
        2.7779023160692304,
        2.7779613470193,
        2.7782446609344333,
        2.837523588910699,
        2.8375828799325973,
        2.8376138389576226,
        2.8974151010625064,
        2.8974736018572003,
        2.8975086980499327,
        2.9570300329942256,
        2.9570731839630753,
        3.0167474569752812,
        3.0168047149199992,
        3.077029956970364,
        3.077089078957215,
        3.0771152179222554,
        3.07715130597353,
        3.077169339871034,
        3.0772005189210176,
        3.1369623269420117,
        3.1370396930724382,
        3.137066813884303,
        3.1371038740035146,
        3.137122980086133,
        3.137154067866504,
        3.197020853869617,
        3.197081048041582,
        3.197105925064534,
        3.197148364968598,
        3.2570217030588537,
        3.257082107011229,
        3.2571172730531543,
        3.316896563861519,
        3.3169559659436345,
        3.3170216199941933,
        3.317081081913784,
        3.376717052888125,
        3.3767882368993014,
        3.3768221710342914,
        3.437399776885286,
        3.4374375890474766,
        3.4374862099066377,
        3.4375081618782133,
        3.4375290509779006,
        3.4375634959433228,
        3.4978361760731786,
        3.497872744919732,
        3.4978964398615062,
        3.497916336869821,
        3.497934941900894,
        3.4979522549547255,
        3.5593524610158056,
        3.559424868086353,
        3.612531960941851,
        3.6125689710024744,
        3.6542840409092605
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 3.6544854808598757,
      "first_text_epoch_s": 1783953537.7284696,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.489649928873405,
      "prompt_id": "risk-register",
      "prompt_index": 9,
      "prompt_sha256": "2dcb017f7bb740ae3a8205fcb464068fcfe82189d61be31850d5e6c6492354bf",
      "prompt_tokens": 61,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953540.2181184,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-09-risk-register",
      "request_started_epoch_s": 1783953536.5636332,
      "response_id_first": "chatcmpl-T1A2Sqkf3v10LchtqsLGxoVsVX18TxKi",
      "response_id_last": "chatcmpl-T1A2Sqkf3v10LchtqsLGxoVsVX18TxKi",
      "response_x_request_id": null,
      "sha256": "d2a92ef0bbbefec1cc0182aec8eba3a45ce56d390f64a3d223947e82f49f40ea",
      "stream_token_id_count": 0,
      "text_preview": "Switching a production inference endpoint runtime (e.g., moving from TensorFlow Serving to Triton Inference Server, or from a custom Flask app to a high-performance C++/Rust service) is a high-stakes operational change. The risks below focus on **model fidelity, latency, resource utilization, and operational stability*",
      "tok_s_1_100_after_ttft": 49.20605118527643,
      "tok_s_after_ttft_full": 51.41285066447935,
      "tok_s_wall_full": 35.025450414399366,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.1648355519864708,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 61,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 189
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.1613612740766257,
        1.2228943491354585,
        1.2229220811277628,
        1.2229362779762596,
        1.2828239020891488,
        1.3437845390290022,
        1.3438188331201673,
        1.3438338320702314,
        1.343844632152468,
        1.3438568550627679,
        1.4034031659830362,
        1.4637853221502155,
        1.463804268045351,
        1.4638181941118091,
        1.4638301960658282,
        1.4638415381778032,
        1.5235197371803224,
        1.523539514048025,
        1.5833678671624511,
        1.583398424088955,
        1.5834113790187985,
        1.643285616999492,
        1.6433175781276077,
        1.6433318650815636,
        1.702332817018032,
        1.7617729580961168,
        1.7618039960507303,
        1.8211670911405236,
        1.8807728730607778,
        1.9401046291459352,
        2.0004091281443834,
        2.0004395351279527,
        2.000454173190519,
        2.0004758841823786,
        2.0607484520878643,
        2.0607815152034163,
        2.0607952810823917,
        2.0608151480555534,
        2.120638000080362,
        2.120671221986413,
        2.1206859201192856,
        2.120706147979945,
        2.1801807831507176,
        2.180212162202224,
        2.2399012811947614,
        2.2995187751948833,
        2.2995491521432996,
        2.3591282041743398,
        2.3591595531906933,
        2.4186968861613423,
        2.4187278649769723,
        2.478095357073471,
        2.5376613850239664,
        2.5973716541193426,
        2.5974155561998487,
        2.6567662581801414,
        2.7164173151832074,
        2.716496985172853,
        2.7765240711160004,
        2.776555721182376,
        2.7765703790355474,
        2.7765911081805825,
        2.836196558084339,
        2.836228067986667,
        2.8958630941342562,
        2.956354425055906,
        3.017506771022454,
        3.0175285020377487,
        3.017543340101838,
        3.0175552130676806,
        3.0175662441179156,
        3.017576623009518,
        3.07747352309525,
        3.077493441058323,
        3.1372009641490877,
        3.197008947143331,
        3.256609478034079,
        3.2566422501113266,
        3.3165545891970396,
        3.316587030189112,
        3.316600495018065,
        3.31662126397714,
        3.3765850111376494,
        3.376617101021111,
        3.3766311979852617,
        3.3766520970966667,
        3.3766632380429655,
        3.4363569649867713,
        3.4364074310287833,
        3.495902353199199,
        3.556149253156036,
        3.5561700521502644,
        3.55618298612535,
        3.556196251185611,
        3.61603426001966,
        3.6160670020617545,
        3.6160806079860777,
        3.6161016770638525,
        3.6760858721099794,
        3.6761174821294844,
        3.676131969084963,
        3.6761530390940607,
        3.676163568161428,
        3.676181351998821,
        3.735899265157059,
        3.796197621151805,
        3.796229331055656,
        3.796243297168985,
        3.7962640770711005,
        3.796276699984446,
        3.796294162981212,
        3.85574457119219,
        3.85577553906478,
        3.9155753860250115,
        3.9748796781059355,
        4.035281439078972,
        4.035313690081239,
        4.035327296005562,
        4.035347584169358,
        4.035358555149287,
        4.035387609153986,
        4.09507755911909,
        4.095118246041238,
        4.095132042188197,
        4.151910403044894,
        4.151948444079608,
        4.202203049091622,
        4.202224349137396
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 4.2023202599957585,
      "first_text_epoch_s": 1783953541.3795927,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 3.040958985919133,
      "prompt_id": "performance-hypotheses",
      "prompt_index": 10,
      "prompt_sha256": "07d3df497692ad49145544fc0388001879fe6a7201aca3f9eaa4e14288f0bfeb",
      "prompt_tokens": 64,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953544.4205503,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-10-performance-hypotheses",
      "request_started_epoch_s": 1783953540.21823,
      "response_id_first": "chatcmpl-U2hdoeVHzGwCvqDVE3IR9697RFqC63me",
      "response_id_last": "chatcmpl-U2hdoeVHzGwCvqDVE3IR9697RFqC63me",
      "response_x_request_id": null,
      "sha256": "60b30326078bb0cd0626be0088274dcb0f91aa06418d90976f5eca55839aee11",
      "stream_token_id_count": 0,
      "text_preview": "Based on the profile indicating that **MoE Gate, Up, and Down projections** are the dominant bottlenecks during the **decode phase**, here is a structured set of performance hypotheses to test.\n\n> **Context:** In MoE decode, the \"Gate\" selects experts, \"Up\" projects input to expert hidden size, and \"Down\" projects back",
      "tok_s_1_100_after_ttft": 39.76528606621022,
      "tok_s_after_ttft_full": 42.09198499311949,
      "tok_s_wall_full": 30.459363418467586,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.1613612740766257,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 64,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 192
      }
    },
    {
      "cached_tokens": 0,
      "chunk_count": 128,
      "chunk_count_equals_completion_tokens": true,
      "chunk_offsets_s": [
        1.159085248131305,
        1.2203380221035331,
        1.2797612380236387,
        1.2798148591537029,
        1.3398206532001495,
        1.3398692850023508,
        1.339895904995501,
        1.3399245191831142,
        1.3399862251244485,
        1.3400141280144453,
        1.3992682851385325,
        1.4590759861748666,
        1.459353709127754,
        1.4593934239819646,
        1.4594087731093168,
        1.5189696191810071,
        1.5190075701102614,
        1.519022447988391,
        1.5791378791909665,
        1.5791679250542074,
        1.5791810501832515,
        1.5792030210141093,
        1.5792137722019106,
        1.5792310140095651,
        1.6387915580999106,
        1.638823859160766,
        1.638837605016306,
        1.638858214020729,
        1.6984930781181902,
        1.6985254790633917,
        1.7578968070447445,
        1.7579281760845333,
        1.8191040339879692,
        1.819143599132076,
        1.8191579959820956,
        1.8191711010877043,
        1.8191815000027418,
        1.819208742119372,
        1.8791189051698893,
        1.8791401451453567,
        1.8791530490852892,
        1.9393252260051668,
        1.939354281174019,
        1.939367335056886,
        1.93938750308007,
        1.9393973420374095,
        1.9394132820889354,
        1.9994575080927461,
        1.9994890470989048,
        1.9995027531404048,
        1.9995231011416763,
        2.05956217716448,
        2.0595943180378526,
        2.0596090760082006,
        2.059630636125803,
        2.059641637140885,
        2.059659270104021,
        2.1192451419774443,
        2.1192802879959345,
        2.1785694411955774,
        2.1786007701884955,
        2.238361060153693,
        2.2981162921059877,
        2.3581496470142156,
        2.358169523999095,
        2.358183840988204,
        2.3581957740243524,
        2.4175753770396113,
        2.4176061651669443,
        2.476854029111564,
        2.537221515085548,
        2.5372535951901227,
        2.5372668001800776,
        2.5372858461923897,
        2.537296255119145,
        2.5373149211518466,
        2.597185248043388,
        2.5972188420128077,
        2.597231245134026,
        2.597250992199406,
        2.597262714058161,
        2.656810874119401,
        2.656843295088038,
        2.6568571410607547,
        2.7163505679927766,
        2.776505272137001,
        2.7765376530587673,
        2.7765517490915954,
        2.7765714460983872,
        2.836875190027058,
        2.8369070100598037,
        2.836921518202871,
        2.8369670030660927,
        2.896305507980287,
        2.9564084140583873,
        2.9564411561004817,
        2.9564558141864836,
        2.956475891172886,
        3.0161166260950267,
        3.0161464321427047,
        3.0760745680890977,
        3.0761059471406043,
        3.076120044104755,
        3.135848534060642,
        3.1358786711934954,
        3.195554291130975,
        3.1955853800754994,
        3.2561134870629758,
        3.256146128056571,
        3.2561609961558133,
        3.256180944154039,
        3.256191423162818,
        3.256208756007254,
        3.3159582260996103,
        3.3159987221006304,
        3.376166891073808,
        3.3761981499847025,
        3.3762127871159464,
        3.4354489489924163,
        3.495424375170842,
        3.4954721750691533,
        3.495495418086648,
        3.5551156841684133,
        3.5551459521520883,
        3.6122481510974467,
        3.6122693410143256,
        3.6122849700041115,
        3.612297003157437
      ],
      "completion_tokens": 128,
      "content_delta_count": 128,
      "elapsed_s": 3.612386532127857,
      "first_text_epoch_s": 1783953545.5797544,
      "metric_chunk_events_at_least_window": true,
      "metric_token_id_events_at_least_window": false,
      "post_ttft_s": 2.4533012839965522,
      "prompt_id": "decision-memo",
      "prompt_index": 11,
      "prompt_sha256": "e1fcb91edf9a4e6048cd5a5738cebcc0f71edc15281607b795bf2a6dddd6adeb",
      "prompt_tokens": 60,
      "reasoning_delta_count": 0,
      "request_ended_epoch_s": 1783953548.0330544,
      "request_id": "bench-qwen36-27b-autoround-int4-b70-realistic-v1-11-decision-memo",
      "request_started_epoch_s": 1783953544.420668,
      "response_id_first": "chatcmpl-UB3FPkiyzkGDi1XptQhzDpujhDxB9exA",
      "response_id_last": "chatcmpl-UB3FPkiyzkGDi1XptQhzDpujhDxB9exA",
      "response_x_request_id": null,
      "sha256": "1b6bca2916680c614e32a77c393698ee91683f6b3772dd6bd207a9e9d1d32bc9",
      "stream_token_id_count": 0,
      "text_preview": "**MEMORANDUM**\n\n**TO:** Engineering Leadership, Product Management, Data Science Team\n**FROM:** [Your Name/Title], Head of Machine Learning\n**DATE:** October 26, 2023\n**SUBJECT:** Decision Memo: Continue Optimization of Current Model (Model A) vs. Pivot to New Model Family (Model B)\n\n### 1. Executive Summary\n\nAfter a c",
      "tok_s_1_100_after_ttft": 53.8485219878389,
      "tok_s_after_ttft_full": 52.17459463090547,
      "tok_s_wall_full": 35.43363891477092,
      "token_id_offsets_s": [],
      "token_timing_source": "openai_stream_content_or_reasoning_delta",
      "ttft_s": 1.159085248131305,
      "usage": {
        "completion_tokens": 128,
        "prompt_tokens": 60,
        "prompt_tokens_details": {
          "cached_tokens": 0
        },
        "total_tokens": 188
      }
    }
  ],
  "run_identity": {
    "api_mode": "chat",
    "base_url": "http://127.0.0.1:19446",
    "created_at_utc": "2026-07-13T14:39:08.033438+00:00",
    "max_tokens": 128,
    "model": "qwen36-27b-mtp-gguf-q4_0",
    "prompt_count": 12,
    "request_extra": {
      "cache_prompt": false
    },
    "return_token_ids": false,
    "seed": 1,
    "suite": {
      "description": "Fixed cold-response prompt suite for Qwen3.6 27B INT4 AutoRound B70 promotion and LocalMaxxing submissions. Each prompt is sent exactly once. Synthetic/repeated prompts remain diagnostic only.",
      "metric": "median generated-token throughput for tokens 1-100 after TTFT across prompts",
      "suite_id": "qwen36-27b-autoround-int4-b70-realistic-v1",
      "version": 1
    },
    "suite_path": "repro/qwen36-27b-autoround-int4-b70/realistic-suite-v1.json"
  },
  "summary": {
    "tok_s_1_100_after_ttft": {
      "count": 12,
      "max": 55.790031389005726,
      "mean": 46.63864668871475,
      "median": 47.81881802978528,
      "min": 36.47145565424477,
      "p10": 39.86953379449598,
      "stdev": 6.246458007352302
    },
    "tok_s_after_ttft_full": {
      "count": 12,
      "max": 52.25037153582198,
      "mean": 47.21721915202124,
      "median": 47.87859819774357,
      "min": 40.536544939571,
      "p10": 41.44464664205012,
      "stdev": 4.371554368715586
    },
    "tok_s_wall_full": {
      "count": 12,
      "max": 35.545980237894995,
      "mean": 32.89298761555302,
      "median": 32.923315078324215,
      "min": 29.373866701044726,
      "p10": 30.164980474029182,
      "stdev": 2.247713682374591
    },
    "ttft_ms": {
      "count": 12,
      "max": 1343.417467083782,
      "mean": 1175.4146496920537,
      "median": 1156.511059962213,
      "min": 1151.226415997371,
      "p10": 1151.996413152665,
      "stdev": 54.53484440139894
    }
  }
}