{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-13T23:26:35Z",
 "ollaya": "0.13.0",
 "machine": "rtx-5090",
 "machine_detail": {
  "id": "rtx-5090",
  "cpu": "AMD Ryzen Threadripper 3960X 24-Core Processor",
  "threads": 48,
  "gpus": [
   {
    "name": "NVIDIA GeForce RTX 5090",
    "driver": "596.21",
    "memory": "32607 MiB"
   }
  ],
  "os": "Linux 6.6.87.2-microsoft-standard-WSL2"
 },
 "device": "cuda",
 "protocol": {
  "preset": "triage",
  "questions": 5,
  "states": 30,
  "warm": 5,
  "n": 20,
  "client": "one request at a time over HTTP to /api/decide, timed by the client; each request a different short message, in turn",
  "memory": "vram_mb: peak rise of the GPU's used memory over its idle level before the load (nvidia-smi, 100 ms); ram_peak_mb and ram_mb: the runner's VmHWM and VmRSS"
 },
 "note": "Edition 2026.10. Ollaya 0.13.0 from the release install script, one model loaded at a time. arbiter:4b and qwen3guard:0.6b were timed again at the end of the session (2026-10-14T06:48:22Z): the first pass had no question set arbiter takes and no cold load for qwen3guard.",
 "results": [
  {
   "model": "decima:small",
   "questions": "triage",
   "load_ms": 4142.7,
   "first_request_ms": 5001.9,
   "answered_by": "decima:small",
   "p50_ms": 13.1,
   "p90_ms": 14.7,
   "min_ms": 12.5,
   "eval_p50_ms": 12.3,
   "input_tokens": 567,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "d0cdad8529bb48df9511afb3f16e97254bceb40d9ec38a6dd4064930e2352db3",
   "size_mb": 486,
   "ram_peak_mb": 1209,
   "ram_mb": 1128,
   "vram_mb": 1270,
   "gpu_idle_mb": 306
  },
  {
   "model": "decima:base",
   "questions": "triage",
   "load_ms": 4931.3,
   "first_request_ms": 4958.6,
   "answered_by": "decima:base",
   "p50_ms": 25.9,
   "p90_ms": 26.3,
   "min_ms": 21.9,
   "eval_p50_ms": 25.1,
   "input_tokens": 458,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "56ad25422fccba646359e24655a82fc836a07b760cdc121aced5a56f222c25f1",
   "size_mb": 1263,
   "ram_peak_mb": 1792,
   "ram_mb": 1138,
   "vram_mb": 2166,
   "gpu_idle_mb": 306
  },
  {
   "model": "decima:agent",
   "questions": "triage",
   "load_ms": 5934.9,
   "first_request_ms": 5960.3,
   "answered_by": "decima:agent",
   "p50_ms": 27.1,
   "p90_ms": 27.6,
   "min_ms": 23.2,
   "eval_p50_ms": 26.2,
   "input_tokens": 463,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "6b5f4883a460936fd5dd3b427fcd57b6711b668495bbdc5c960f5fbb62532c83",
   "size_mb": 1263,
   "ram_peak_mb": 1823,
   "ram_mb": 1170,
   "vram_mb": 2806,
   "gpu_idle_mb": 306
  },
  {
   "model": "laya:en",
   "questions": "triage",
   "load_ms": 3833.4,
   "first_request_ms": 3864.4,
   "answered_by": "laya:en",
   "p50_ms": 18.3,
   "p90_ms": 19.2,
   "min_ms": 15.2,
   "eval_p50_ms": 17.4,
   "input_tokens": 349,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "c305a9276531a47000bf93559d2c94f1ed6cbb67055c9151084682dad7655e9d",
   "size_mb": 813,
   "ram_peak_mb": 1266,
   "ram_mb": 1266,
   "vram_mb": 1654,
   "gpu_idle_mb": 306
  },
  {
   "model": "laya:multilingual",
   "questions": "triage",
   "load_ms": 3619.3,
   "first_request_ms": 3649.5,
   "answered_by": "laya:multilingual",
   "p50_ms": 14.3,
   "p90_ms": 14.6,
   "min_ms": 13.5,
   "eval_p50_ms": 13.5,
   "input_tokens": 356,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "2840506e1f978aeb696a6f84af43ecc7fca95754cb63a67bafe6132532d384eb",
   "size_mb": 652,
   "ram_peak_mb": 1204,
   "ram_mb": 1034,
   "vram_mb": 1398,
   "gpu_idle_mb": 306
  },
  {
   "model": "laya:typed-decisions",
   "questions": "triage",
   "load_ms": 3862.5,
   "first_request_ms": 3900.9,
   "answered_by": "laya:typed-decisions",
   "p50_ms": 20.4,
   "p90_ms": 21.1,
   "min_ms": 14.5,
   "eval_p50_ms": 19.6,
   "input_tokens": 344,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "6d17e5fbcbb8215a74f2efd0dcc0ffdbd472e9b25ccb07867f9f146a791673ba",
   "size_mb": 813,
   "ram_peak_mb": 1331,
   "ram_mb": 1281,
   "vram_mb": 1654,
   "gpu_idle_mb": 306
  },
  {
   "model": "nli:deberta-v3-large",
   "questions": "triage",
   "load_ms": 8222.5,
   "first_request_ms": 8242.5,
   "answered_by": "nli:deberta-v3-large",
   "p50_ms": 17.7,
   "p90_ms": 18,
   "min_ms": 17.4,
   "eval_p50_ms": 16.8,
   "input_tokens": 619,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "b33142f79a0338ab5398c7be66c18f58483c07aaec8262dc1f314aa02cbee205",
   "size_mb": 843,
   "ram_peak_mb": 4103,
   "ram_mb": 3557,
   "vram_mb": 2682,
   "gpu_idle_mb": 306
  },
  {
   "model": "nli:modernbert-large",
   "questions": "triage",
   "load_ms": 6018.1,
   "first_request_ms": 6036.2,
   "answered_by": "nli:modernbert-large",
   "p50_ms": 16.8,
   "p90_ms": 17.2,
   "min_ms": 16.4,
   "eval_p50_ms": 15.9,
   "input_tokens": 408,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "0c984ae95685d511953341fd75e5cf49e31bae5110574dba78999c5f2267cd7f",
   "size_mb": 762,
   "ram_peak_mb": 3329,
   "ram_mb": 3101,
   "vram_mb": 2678,
   "gpu_idle_mb": 306
  },
  {
   "model": "gliclass:large",
   "questions": "triage",
   "load_ms": 7149.2,
   "first_request_ms": 7165.9,
   "answered_by": "gliclass:large",
   "p50_ms": 14.1,
   "p90_ms": 14.5,
   "min_ms": 13.7,
   "eval_p50_ms": 13.3,
   "input_tokens": 273,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "8c4c3acece90b83f44da3694f0aedcb90503de37aff702995e18dc317a63e5a6",
   "size_mb": 1687,
   "ram_peak_mb": 2821,
   "ram_mb": 2375,
   "vram_mb": 2702,
   "gpu_idle_mb": 306
  },
  {
   "model": "von:1.1",
   "questions": "triage",
   "load_ms": 5703.1,
   "first_request_ms": 5723.3,
   "answered_by": "von:1.1",
   "p50_ms": 19.3,
   "p90_ms": 19.8,
   "min_ms": 18.4,
   "eval_p50_ms": 18.4,
   "input_tokens": 379,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "25606bfd644593e63b4c63b8219d4fd0085d11634d6fa4701268d77cf804f689",
   "size_mb": 1514,
   "ram_peak_mb": 2136,
   "ram_mb": 1939,
   "vram_mb": 2686,
   "gpu_idle_mb": 306
  },
  {
   "model": "qwen3guard:0.6b",
   "questions": "builtin (4)",
   "load_ms": 10289.2,
   "first_request_ms": 10320.9,
   "answered_by": "qwen3guard:0.6b",
   "p50_ms": 30.2,
   "p90_ms": 32.3,
   "min_ms": 28.6,
   "eval_p50_ms": 29.1,
   "input_tokens": 642,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "396892330ef0cb0c3057575dee6a9c46a8c8bfe49afafb738ed1addf2c466a63",
   "size_mb": 1450,
   "ram_peak_mb": 5499,
   "ram_mb": 4638,
   "vram_mb": 3701,
   "gpu_idle_mb": 248
  },
  {
   "model": "snap:2b",
   "questions": "triage",
   "load_ms": 6249.7,
   "first_request_ms": 6335.7,
   "answered_by": "snap:2b",
   "p50_ms": 82.3,
   "p90_ms": 84.6,
   "min_ms": 81.3,
   "eval_p50_ms": 81.4,
   "input_tokens": 606,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "f356b13c94415f1237ab94d752b0de1705ed98f38ee389549264dfa53532f171",
   "size_mb": 2556,
   "ram_peak_mb": 2922,
   "ram_mb": 713,
   "vram_mb": 3450,
   "gpu_idle_mb": 306
  },
  {
   "model": "snap:2b-q4_k_m",
   "questions": "triage",
   "load_ms": 4096.6,
   "first_request_ms": 4180.1,
   "answered_by": "snap:2b-q4_k_m",
   "p50_ms": 78.2,
   "p90_ms": 78.7,
   "min_ms": 77.4,
   "eval_p50_ms": 77.3,
   "input_tokens": 606,
   "device": "cuda:0",
   "precision": "Q4_K - MEDIUM",
   "digest": "a19db7952873b56351a1a1e820fa571c65a8f50b4e9862addbf0dda063661036",
   "size_mb": 1489,
   "ram_peak_mb": 1855,
   "ram_mb": 592,
   "vram_mb": 2514,
   "gpu_idle_mb": 306
  },
  {
   "model": "d1:3b",
   "questions": "triage",
   "load_ms": 6859.9,
   "first_request_ms": 6925.4,
   "answered_by": "d1:3b",
   "p50_ms": 50.8,
   "p90_ms": 52.5,
   "min_ms": 50,
   "eval_p50_ms": 49.8,
   "input_tokens": 350,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "d92cc117d0d9ef8624b4408ef4823b4b8efef403fee9f7d54629cb037a2b58da",
   "size_mb": 2742,
   "ram_peak_mb": 3131,
   "ram_mb": 748,
   "vram_mb": 4082,
   "gpu_idle_mb": 306
  },
  {
   "model": "winnow:e4b",
   "questions": "triage",
   "load_ms": 17524.7,
   "first_request_ms": 17648.1,
   "answered_by": "winnow:e4b",
   "p50_ms": 112.6,
   "p90_ms": 119.4,
   "min_ms": 107.7,
   "eval_p50_ms": 111.7,
   "input_tokens": 696,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "dd4bf88aa50bebb02e7a26fbebed14682befd037e3e89789ee22b9793f82d226",
   "size_mb": 7635,
   "ram_peak_mb": 8079,
   "ram_mb": 4092,
   "vram_mb": 6356,
   "gpu_idle_mb": 306
  },
  {
   "model": "credence:e4b",
   "questions": "triage",
   "load_ms": 17863,
   "first_request_ms": 17986.8,
   "answered_by": "credence:e4b",
   "p50_ms": 120.7,
   "p90_ms": 122.5,
   "min_ms": 118.3,
   "eval_p50_ms": 119.7,
   "input_tokens": 691,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "ab2f84b8484681a88d465b7771f43804886e7bf4d529a731d4b8ce24bea57b69",
   "size_mb": 8315,
   "ram_peak_mb": 8759,
   "ram_mb": 4093,
   "vram_mb": 6356,
   "gpu_idle_mb": 306
  },
  {
   "model": "credence:e4b-calibrated",
   "questions": "triage",
   "load_ms": 17368.1,
   "first_request_ms": 17482.5,
   "answered_by": "credence:e4b-calibrated",
   "p50_ms": 114.3,
   "p90_ms": 115.4,
   "min_ms": 110.4,
   "eval_p50_ms": 113.4,
   "input_tokens": 711,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "e9d941a002f0bba8438d09f8c8fe9bdad8a830dab0ea8cba13cd758e1fafb565",
   "size_mb": 8315,
   "ram_peak_mb": 8759,
   "ram_mb": 4093,
   "vram_mb": 6356,
   "gpu_idle_mb": 306
  },
  {
   "model": "jevk5:4b",
   "questions": "triage",
   "load_ms": 10392.8,
   "first_request_ms": 10491.2,
   "answered_by": "jevk5:4b",
   "p50_ms": 95.3,
   "p90_ms": 97.5,
   "min_ms": 94.3,
   "eval_p50_ms": 94.5,
   "input_tokens": 752,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "58b5d6516410070d5f165d77357ed7bfefe5ffedde63e6ae75ab5101d4fe0221",
   "size_mb": 4275,
   "ram_peak_mb": 4686,
   "ram_mb": 1152,
   "vram_mb": 5910,
   "gpu_idle_mb": 306
  },
  {
   "model": "decider:0.8b",
   "questions": "triage",
   "load_ms": 17435.7,
   "first_request_ms": 17705.1,
   "answered_by": "decider:0.8b",
   "p50_ms": 189,
   "p90_ms": 190.8,
   "min_ms": 173.9,
   "eval_p50_ms": 187.8,
   "input_tokens": 544,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "97a22ac9c6c42eebb762ddb5eb47f0542b5e0788cf1cea31cee6205d5bd1064a",
   "size_mb": 1462,
   "ram_peak_mb": 7023,
   "ram_mb": 5661,
   "vram_mb": 5116,
   "gpu_idle_mb": 306
  },
  {
   "model": "decider:2b",
   "questions": "triage",
   "load_ms": 32189.4,
   "first_request_ms": 32500.7,
   "answered_by": "decider:2b",
   "p50_ms": 192.7,
   "p90_ms": 194.1,
   "min_ms": 189.8,
   "eval_p50_ms": 191.7,
   "input_tokens": 520,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "5507d6f80e9a2c82d7dd4c0eb6cffc03916669a4ce5f0fe506ab5e434cf9a62a",
   "size_mb": 3616,
   "ram_peak_mb": 11706,
   "ram_mb": 1445,
   "vram_mb": 12176,
   "gpu_idle_mb": 306
  },
  {
   "model": "decider:4b",
   "questions": "triage",
   "load_ms": 22062.3,
   "first_request_ms": 22613.1,
   "answered_by": "decider:4b",
   "p50_ms": 393,
   "p90_ms": 393.9,
   "min_ms": 392.1,
   "eval_p50_ms": 392.1,
   "input_tokens": 544,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "920dea4d1851680da7129db890b97542e824a3709f2ccb09b7f35ce6b4ff4669",
   "size_mb": 8051,
   "ram_peak_mb": 2008,
   "ram_mb": 1416,
   "vram_mb": 11278,
   "gpu_idle_mb": 306
  },
  {
   "model": "decision:eos",
   "questions": "triage",
   "load_ms": 13277,
   "first_request_ms": 13590.8,
   "answered_by": "decision:eos",
   "p50_ms": 209.2,
   "p90_ms": 210.2,
   "min_ms": 208.5,
   "eval_p50_ms": 208.3,
   "input_tokens": 680,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "fc119d7067ecd029a5316442cf42a394236fff8758ab296f3d25a1285a34e047",
   "size_mb": 1466,
   "ram_peak_mb": 6095,
   "ram_mb": 5717,
   "vram_mb": 5910,
   "gpu_idle_mb": 306
  },
  {
   "model": "kev:0.8b",
   "questions": "triage",
   "load_ms": 11888.6,
   "first_request_ms": 12093.1,
   "answered_by": "kev:0.8b",
   "p50_ms": 141.8,
   "p90_ms": 142.7,
   "min_ms": 141.3,
   "eval_p50_ms": 140.8,
   "input_tokens": 283,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "8a0d4094df504ca7329bc999dabbb89c15ad5221f55728acde0ff61c2b3909f2",
   "size_mb": 1738,
   "ram_peak_mb": 5371,
   "ram_mb": 5334,
   "vram_mb": 4650,
   "gpu_idle_mb": 306
  },
  {
   "model": "kev:4b",
   "questions": "triage",
   "load_ms": 21586,
   "first_request_ms": 21977.1,
   "answered_by": "kev:4b",
   "p50_ms": 318.7,
   "p90_ms": 320.1,
   "min_ms": 300,
   "eval_p50_ms": 317.7,
   "input_tokens": 283,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "c0e732655dc9d799dd21ba7f2f9a445fbd50ee83578e96ccfbf691c1be9dde53",
   "size_mb": 9050,
   "ram_peak_mb": 2073,
   "ram_mb": 1524,
   "vram_mb": 11200,
   "gpu_idle_mb": 306
  },
  {
   "model": "kev:9b",
   "questions": "triage",
   "load_ms": 36583.3,
   "first_request_ms": 37104.8,
   "answered_by": "kev:9b",
   "p50_ms": 374.8,
   "p90_ms": 375.3,
   "min_ms": 373.6,
   "eval_p50_ms": 373.8,
   "input_tokens": 283,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "c9130c1a1e629c13497c833300395b56c7e347ecd3558596a2c2e1ce41ce7eda",
   "size_mb": 18618,
   "ram_peak_mb": 2816,
   "ram_mb": 1781,
   "vram_mb": 18438,
   "gpu_idle_mb": 306
  },
  {
   "model": "arbiter:4b",
   "questions": "triage, frustration on six levels",
   "load_ms": 6789.5,
   "first_request_ms": 6930.2,
   "answered_by": "arbiter:4b",
   "p50_ms": 134.5,
   "p90_ms": 134.7,
   "min_ms": 133.9,
   "eval_p50_ms": 133.4,
   "input_tokens": 350,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "2896b8aecfca38e37f22e75e78dca0a6955eba6cfd8dc5b24b637eb124b8cff8",
   "size_mb": 8370,
   "ram_peak_mb": 2044,
   "ram_mb": 1168,
   "vram_mb": 8449,
   "gpu_idle_mb": 248
  },
  {
   "model": "winnow:12b",
   "questions": "triage",
   "load_ms": 27609.2,
   "first_request_ms": 27773.7,
   "answered_by": "winnow:12b",
   "p50_ms": 137.5,
   "p90_ms": 149.1,
   "min_ms": 135.9,
   "eval_p50_ms": 136.6,
   "input_tokens": 706,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "8d72fee0e5f92a625cdb85ae4f6abdc024c9e6f124a0ca67e64522c3f24c47ce",
   "size_mb": 12083,
   "ram_peak_mb": 12526,
   "ram_mb": 1562,
   "vram_mb": 15872,
   "gpu_idle_mb": 306
  },
  {
   "model": "cygnet:12b",
   "questions": "triage",
   "load_ms": 28013.8,
   "first_request_ms": 28177.6,
   "answered_by": "cygnet:12b",
   "p50_ms": 152.2,
   "p90_ms": 153.1,
   "min_ms": 150.2,
   "eval_p50_ms": 151.2,
   "input_tokens": 746,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "81160d85e13acfda209f78ba38a3e1d7ef0c80727820b022bbd9c878476e93cc",
   "size_mb": 12083,
   "ram_peak_mb": 12509,
   "ram_mb": 1553,
   "vram_mb": 18560,
   "gpu_idle_mb": 306
  },
  {
   "model": "clm:8b",
   "questions": "triage",
   "load_ms": 31936.4,
   "first_request_ms": 32159.5,
   "answered_by": "clm:8b",
   "p50_ms": 94.5,
   "p90_ms": 95.1,
   "min_ms": 93.2,
   "eval_p50_ms": 93.5,
   "input_tokens": 317,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "3f89a66972f1398210d2f41b94667665132aa51c2d313237c3fc70a49c275cb5",
   "size_mb": 15712,
   "ram_peak_mb": 1796,
   "ram_mb": 1283,
   "vram_mb": 15678,
   "gpu_idle_mb": 306
  },
  {
   "model": "jeb:4b",
   "questions": "triage",
   "load_ms": 10487.9,
   "first_request_ms": 10581.8,
   "answered_by": "jeb:4b",
   "p50_ms": 90.7,
   "p90_ms": 92.4,
   "min_ms": 88.4,
   "eval_p50_ms": 89.8,
   "input_tokens": 564,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "4aba27707f6f4ba6f116710a0d05743220e5a5819b8826a9bd32385d0d9df2ff",
   "size_mb": 4397,
   "ram_peak_mb": 4808,
   "ram_mb": 1139,
   "vram_mb": 5530,
   "gpu_idle_mb": 306
  },
  {
   "model": "jeb:9b",
   "questions": "triage",
   "load_ms": 20756.8,
   "first_request_ms": 20881.2,
   "answered_by": "jeb:9b",
   "p50_ms": 106.9,
   "p90_ms": 107.7,
   "min_ms": 105.4,
   "eval_p50_ms": 105.9,
   "input_tokens": 569,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "dfe0411a8d060f1fa64ff2539b1e4b44a061c11cce4d5d96a82a7077b01d8b9e",
   "size_mb": 9333,
   "ram_peak_mb": 9744,
   "ram_mb": 1532,
   "vram_mb": 9316,
   "gpu_idle_mb": 306
  },
  {
   "model": "clef:flash",
   "questions": "triage",
   "load_ms": 38224.8,
   "first_request_ms": 38706.2,
   "answered_by": "clef:flash",
   "p50_ms": 406.4,
   "p90_ms": 407.2,
   "min_ms": 403.3,
   "eval_p50_ms": 405.4,
   "input_tokens": 632,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "acda6b87d74dc55e04916220fcfc643ca2d409ca22ec5e9d18bf53b91604845d",
   "size_mb": 18211,
   "ram_peak_mb": 3040,
   "ram_mb": 1729,
   "vram_mb": 20244,
   "gpu_idle_mb": 306
  },
  {
   "model": "jeeves:9b",
   "questions": "triage",
   "load_ms": 35354.7,
   "first_request_ms": 36151.5,
   "answered_by": "jeeves:9b",
   "p50_ms": 641.1,
   "p90_ms": 643,
   "min_ms": 636.3,
   "eval_p50_ms": 640,
   "input_tokens": 498,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "0f4a1ff8396c0718c432603d96c0a8eb975a0d3bab83bd374bd35e36dfc5512d",
   "size_mb": 17116,
   "ram_peak_mb": 2720,
   "ram_mb": 1719,
   "vram_mb": 18528,
   "gpu_idle_mb": 306
  },
  {
   "model": "nimble:9b",
   "questions": "triage",
   "load_ms": 39745.3,
   "first_request_ms": 41576.8,
   "answered_by": "nimble:9b",
   "p50_ms": 1688.1,
   "p90_ms": 1694.1,
   "min_ms": 1682.5,
   "eval_p50_ms": 1687.1,
   "input_tokens": 2973,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "fb7e7f742e34636e853f91c467b9463e3155dbb08cabc5c49906d52a5750487d",
   "size_mb": 18610,
   "ram_peak_mb": 3086,
   "ram_mb": 1940,
   "vram_mb": 20782,
   "gpu_idle_mb": 306
  },
  {
   "model": "jeb:27b",
   "questions": "triage",
   "load_ms": 36285.8,
   "first_request_ms": 36539.9,
   "answered_by": "jeb:27b",
   "p50_ms": 252.6,
   "p90_ms": 254.4,
   "min_ms": 243.1,
   "eval_p50_ms": 251.7,
   "input_tokens": 564,
   "device": "cuda:0",
   "precision": "Q4_K - MEDIUM",
   "digest": "da59bd8b0f90123b7ca0de07f2a325aa860b8ab97814ba36226f11463186f561",
   "size_mb": 16032,
   "ram_peak_mb": 16444,
   "ram_mb": 1206,
   "vram_mb": 16592,
   "gpu_idle_mb": 306
  }
 ],
 "edition": "2026.10"
}