{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-14T05:37:16Z",
 "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": "cpu",
 "protocol": {
  "preset": "triage",
  "questions": 5,
  "states": 30,
  "warm": 2,
  "n": 10,
  "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 with OLLAYA_DEVICE=cpu, one model loaded at a time.",
 "results": [
  {
   "model": "decima:small",
   "questions": "triage",
   "load_ms": 1812.6,
   "first_request_ms": 2503.3,
   "answered_by": "decima:small",
   "p50_ms": 625,
   "p90_ms": 645.1,
   "min_ms": 611.3,
   "eval_p50_ms": 624,
   "input_tokens": 567,
   "device": "cpu",
   "precision": "F32",
   "digest": "d0cdad8529bb48df9511afb3f16e97254bceb40d9ec38a6dd4064930e2352db3",
   "size_mb": 486,
   "ram_peak_mb": 589,
   "ram_mb": 589
  },
  {
   "model": "decima:base",
   "questions": "triage",
   "load_ms": 4143.3,
   "first_request_ms": 5452.7,
   "answered_by": "decima:base",
   "p50_ms": 1210.6,
   "p90_ms": 1257.1,
   "min_ms": 1167.2,
   "eval_p50_ms": 1209.6,
   "input_tokens": 458,
   "device": "cpu",
   "precision": "F32",
   "digest": "56ad25422fccba646359e24655a82fc836a07b760cdc121aced5a56f222c25f1",
   "size_mb": 1263,
   "ram_peak_mb": 1125,
   "ram_mb": 1125
  },
  {
   "model": "decima:agent",
   "questions": "triage",
   "load_ms": 4158.2,
   "first_request_ms": 5449.2,
   "answered_by": "decima:agent",
   "p50_ms": 1230.2,
   "p90_ms": 1267.6,
   "min_ms": 1180.2,
   "eval_p50_ms": 1229.1,
   "input_tokens": 458,
   "device": "cpu",
   "precision": "F32",
   "digest": "6b5f4883a460936fd5dd3b427fcd57b6711b668495bbdc5c960f5fbb62532c83",
   "size_mb": 1263,
   "ram_peak_mb": 1131,
   "ram_mb": 1131
  },
  {
   "model": "laya:en",
   "questions": "triage",
   "load_ms": 6337.7,
   "first_request_ms": 7293.1,
   "answered_by": "laya:en",
   "p50_ms": 945.2,
   "p90_ms": 950.2,
   "min_ms": 918.4,
   "eval_p50_ms": 944.1,
   "input_tokens": 349,
   "device": "cpu",
   "precision": "F32",
   "digest": "c305a9276531a47000bf93559d2c94f1ed6cbb67055c9151084682dad7655e9d",
   "size_mb": 813,
   "ram_peak_mb": 3207,
   "ram_mb": 3167
  },
  {
   "model": "laya:multilingual",
   "questions": "triage",
   "load_ms": 4780.1,
   "first_request_ms": 5295.1,
   "answered_by": "laya:multilingual",
   "p50_ms": 515.5,
   "p90_ms": 517.2,
   "min_ms": 508.3,
   "eval_p50_ms": 514.5,
   "input_tokens": 356,
   "device": "cpu",
   "precision": "F32",
   "digest": "2840506e1f978aeb696a6f84af43ecc7fca95754cb63a67bafe6132532d384eb",
   "size_mb": 652,
   "ram_peak_mb": 1941,
   "ram_mb": 1927
  },
  {
   "model": "laya:typed-decisions",
   "questions": "triage",
   "load_ms": 6842.3,
   "first_request_ms": 7775.7,
   "answered_by": "laya:typed-decisions",
   "p50_ms": 944.1,
   "p90_ms": 952,
   "min_ms": 923.1,
   "eval_p50_ms": 942.8,
   "input_tokens": 339,
   "device": "cpu",
   "precision": "F32",
   "digest": "6d17e5fbcbb8215a74f2efd0dcc0ffdbd472e9b25ccb07867f9f146a791673ba",
   "size_mb": 813,
   "ram_peak_mb": 3215,
   "ram_mb": 3215
  },
  {
   "model": "nli:deberta-v3-large",
   "questions": "triage",
   "load_ms": 7709.9,
   "first_request_ms": 10086.7,
   "answered_by": "nli:deberta-v3-large",
   "p50_ms": 2318.2,
   "p90_ms": 2438.6,
   "min_ms": 2052,
   "eval_p50_ms": 2317.2,
   "input_tokens": 580,
   "device": "cpu",
   "precision": "F32",
   "digest": "b33142f79a0338ab5398c7be66c18f58483c07aaec8262dc1f314aa02cbee205",
   "size_mb": 843,
   "ram_peak_mb": 3848,
   "ram_mb": 3848
  },
  {
   "model": "nli:modernbert-large",
   "questions": "triage",
   "load_ms": 6018,
   "first_request_ms": 8116.9,
   "answered_by": "nli:modernbert-large",
   "p50_ms": 2101.4,
   "p90_ms": 2141.9,
   "min_ms": 1889.9,
   "eval_p50_ms": 2100.4,
   "input_tokens": 408,
   "device": "cpu",
   "precision": "F32",
   "digest": "0c984ae95685d511953341fd75e5cf49e31bae5110574dba78999c5f2267cd7f",
   "size_mb": 762,
   "ram_peak_mb": 2991,
   "ram_mb": 2945
  },
  {
   "model": "gliclass:large",
   "questions": "triage",
   "load_ms": 6786.5,
   "first_request_ms": 8647.6,
   "answered_by": "gliclass:large",
   "p50_ms": 2133.2,
   "p90_ms": 2185.3,
   "min_ms": 1595.5,
   "eval_p50_ms": 2132.1,
   "input_tokens": 281,
   "device": "cpu",
   "precision": "F32",
   "digest": "8c4c3acece90b83f44da3694f0aedcb90503de37aff702995e18dc317a63e5a6",
   "size_mb": 1687,
   "ram_peak_mb": 2415,
   "ram_mb": 2415
  },
  {
   "model": "von:1.1",
   "questions": "triage",
   "load_ms": 5948.1,
   "first_request_ms": 8778.4,
   "answered_by": "von:1.1",
   "p50_ms": 2590.6,
   "p90_ms": 2602.9,
   "min_ms": 2237.5,
   "eval_p50_ms": 2589.5,
   "input_tokens": 394,
   "device": "cpu",
   "precision": "F32",
   "digest": "25606bfd644593e63b4c63b8219d4fd0085d11634d6fa4701268d77cf804f689",
   "size_mb": 1514,
   "ram_peak_mb": 1747,
   "ram_mb": 1747
  },
  {
   "model": "qwen3guard:0.6b",
   "questions": "builtin (4)",
   "load_ms": 9624.7,
   "first_request_ms": 11123.1,
   "answered_by": "qwen3guard:0.6b",
   "p50_ms": 1493.7,
   "p90_ms": 1505.5,
   "min_ms": 1486,
   "eval_p50_ms": 1492.7,
   "input_tokens": 642,
   "device": "cpu",
   "precision": "F32",
   "digest": "396892330ef0cb0c3057575dee6a9c46a8c8bfe49afafb738ed1addf2c466a63",
   "size_mb": 1450,
   "ram_peak_mb": 5152,
   "ram_mb": 4732
  },
  {
   "model": "snap:2b",
   "questions": "triage",
   "load_ms": 6428.3,
   "first_request_ms": 8717.1,
   "answered_by": "snap:2b",
   "p50_ms": 2649.8,
   "p90_ms": 2729.4,
   "min_ms": 2462.9,
   "eval_p50_ms": 2648.5,
   "input_tokens": 606,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "f356b13c94415f1237ab94d752b0de1705ed98f38ee389549264dfa53532f171",
   "size_mb": 2556,
   "ram_peak_mb": 2958,
   "ram_mb": 2958
  },
  {
   "model": "snap:2b-q4_k_m",
   "questions": "triage",
   "load_ms": 5514.5,
   "first_request_ms": 7276.8,
   "answered_by": "snap:2b-q4_k_m",
   "p50_ms": 1734.7,
   "p90_ms": 1765,
   "min_ms": 1615.4,
   "eval_p50_ms": 1733.5,
   "input_tokens": 616,
   "device": "cpu",
   "precision": "Q4_K - MEDIUM",
   "digest": "a19db7952873b56351a1a1e820fa571c65a8f50b4e9862addbf0dda063661036",
   "size_mb": 1489,
   "ram_peak_mb": 2802,
   "ram_mb": 2802
  },
  {
   "model": "d1:3b",
   "questions": "triage",
   "load_ms": 6607.3,
   "first_request_ms": 8477.7,
   "answered_by": "d1:3b",
   "p50_ms": 1757.4,
   "p90_ms": 1944.1,
   "min_ms": 1564.2,
   "eval_p50_ms": 1756.1,
   "input_tokens": 350,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "d92cc117d0d9ef8624b4408ef4823b4b8efef403fee9f7d54629cb037a2b58da",
   "size_mb": 2742,
   "ram_peak_mb": 3349,
   "ram_mb": 3349
  },
  {
   "model": "winnow:e4b",
   "questions": "triage",
   "load_ms": 18974.6,
   "first_request_ms": 22991.9,
   "answered_by": "winnow:e4b",
   "p50_ms": 4095.3,
   "p90_ms": 4149.7,
   "min_ms": 3530.1,
   "eval_p50_ms": 4093.9,
   "input_tokens": 696,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "dd4bf88aa50bebb02e7a26fbebed14682befd037e3e89789ee22b9793f82d226",
   "size_mb": 7635,
   "ram_peak_mb": 8265,
   "ram_mb": 8265
  },
  {
   "model": "credence:e4b",
   "questions": "triage",
   "load_ms": 19870.5,
   "first_request_ms": 23474.9,
   "answered_by": "credence:e4b",
   "p50_ms": 4038.9,
   "p90_ms": 4113.5,
   "min_ms": 3411.2,
   "eval_p50_ms": 4037.8,
   "input_tokens": 686,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "ab2f84b8484681a88d465b7771f43804886e7bf4d529a731d4b8ce24bea57b69",
   "size_mb": 8315,
   "ram_peak_mb": 8946,
   "ram_mb": 8946
  },
  {
   "model": "credence:e4b-calibrated",
   "questions": "triage",
   "load_ms": 19453.5,
   "first_request_ms": 23603,
   "answered_by": "credence:e4b-calibrated",
   "p50_ms": 4097.1,
   "p90_ms": 4183.9,
   "min_ms": 3832.2,
   "eval_p50_ms": 4095.9,
   "input_tokens": 691,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "e9d941a002f0bba8438d09f8c8fe9bdad8a830dab0ea8cba13cd758e1fafb565",
   "size_mb": 8315,
   "ram_peak_mb": 8945,
   "ram_mb": 8945
  },
  {
   "model": "jevk5:4b",
   "questions": "triage",
   "load_ms": 11265.7,
   "first_request_ms": 18369.1,
   "answered_by": "jevk5:4b",
   "p50_ms": 6979.7,
   "p90_ms": 7267.8,
   "min_ms": 5952,
   "eval_p50_ms": 6978.6,
   "input_tokens": 757,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "58b5d6516410070d5f165d77357ed7bfefe5ffedde63e6ae75ab5101d4fe0221",
   "size_mb": 4275,
   "ram_peak_mb": 4997,
   "ram_mb": 4997
  },
  {
   "model": "decider:0.8b",
   "questions": "triage",
   "load_ms": 15122.1,
   "first_request_ms": 18494.1,
   "answered_by": "decider:0.8b",
   "p50_ms": 3105.5,
   "p90_ms": 3108.8,
   "min_ms": 3087.3,
   "eval_p50_ms": 3104.6,
   "input_tokens": 568,
   "device": "cpu",
   "precision": "F32",
   "digest": "97a22ac9c6c42eebb762ddb5eb47f0542b5e0788cf1cea31cee6205d5bd1064a",
   "size_mb": 1462,
   "ram_peak_mb": 6740,
   "ram_mb": 6496
  },
  {
   "model": "decider:2b",
   "questions": "triage",
   "load_ms": 36578.8,
   "first_request_ms": 41374.1,
   "answered_by": "decider:2b",
   "p50_ms": 4540.2,
   "p90_ms": 4560.1,
   "min_ms": 4522.1,
   "eval_p50_ms": 4538.8,
   "input_tokens": 544,
   "device": "cpu",
   "precision": "F32",
   "digest": "5507d6f80e9a2c82d7dd4c0eb6cffc03916669a4ce5f0fe506ab5e434cf9a62a",
   "size_mb": 3616,
   "ram_peak_mb": 14924,
   "ram_mb": 14924
  },
  {
   "model": "decider:4b",
   "questions": "triage",
   "load_ms": 20199,
   "first_request_ms": 31179.9,
   "answered_by": "decider:4b",
   "p50_ms": 11875.9,
   "p90_ms": 11918.5,
   "min_ms": 11851.9,
   "eval_p50_ms": 11874.9,
   "input_tokens": 552,
   "device": "cpu",
   "precision": "F32",
   "digest": "920dea4d1851680da7129db890b97542e824a3709f2ccb09b7f35ce6b4ff4669",
   "size_mb": 8051,
   "ram_peak_mb": 9524,
   "ram_mb": 9524
  },
  {
   "model": "decision:eos",
   "questions": "triage",
   "load_ms": 14264.6,
   "first_request_ms": 17932.8,
   "answered_by": "decision:eos",
   "p50_ms": 3505.7,
   "p90_ms": 4296.9,
   "min_ms": 3485.2,
   "eval_p50_ms": 3504.7,
   "input_tokens": 690,
   "device": "cpu",
   "precision": "F32",
   "digest": "fc119d7067ecd029a5316442cf42a394236fff8758ab296f3d25a1285a34e047",
   "size_mb": 1466,
   "ram_peak_mb": 6746,
   "ram_mb": 6746
  },
  {
   "model": "kev:0.8b",
   "questions": "triage",
   "load_ms": 12524.8,
   "first_request_ms": 15723.7,
   "answered_by": "kev:0.8b",
   "p50_ms": 2907.7,
   "p90_ms": 2911.3,
   "min_ms": 2895.2,
   "eval_p50_ms": 2906.6,
   "input_tokens": 283,
   "device": "cpu",
   "precision": "F32",
   "digest": "8a0d4094df504ca7329bc999dabbb89c15ad5221f55728acde0ff61c2b3909f2",
   "size_mb": 1738,
   "ram_peak_mb": 5086,
   "ram_mb": 5086
  },
  {
   "model": "kev:4b",
   "questions": "triage",
   "load_ms": 22526.9,
   "first_request_ms": 31406.5,
   "answered_by": "kev:4b",
   "p50_ms": 8513.1,
   "p90_ms": 8519.9,
   "min_ms": 8451.2,
   "eval_p50_ms": 8512,
   "input_tokens": 288,
   "device": "cpu",
   "precision": "F32",
   "digest": "c0e732655dc9d799dd21ba7f2f9a445fbd50ee83578e96ccfbf691c1be9dde53",
   "size_mb": 9050,
   "ram_peak_mb": 9146,
   "ram_mb": 9146
  },
  {
   "model": "kev:9b",
   "questions": "triage",
   "load_ms": 39177.7,
   "first_request_ms": 52121.7,
   "answered_by": "kev:9b",
   "p50_ms": 12497.5,
   "p90_ms": 12524.1,
   "min_ms": 12442.2,
   "eval_p50_ms": 12496.5,
   "input_tokens": 268,
   "device": "cpu",
   "precision": "F32",
   "digest": "c9130c1a1e629c13497c833300395b56c7e347ecd3558596a2c2e1ce41ce7eda",
   "size_mb": 18618,
   "ram_peak_mb": 16262,
   "ram_mb": 16262
  },
  {
   "model": "arbiter:4b",
   "questions": "triage, frustration on six levels",
   "load_ms": 6571.7,
   "first_request_ms": 12911.2,
   "answered_by": "arbiter:4b",
   "p50_ms": 6110.1,
   "p90_ms": 6130.7,
   "min_ms": 6076.3,
   "eval_p50_ms": 6108.8,
   "input_tokens": 345,
   "device": "cpu",
   "precision": "F32",
   "digest": "2896b8aecfca38e37f22e75e78dca0a6955eba6cfd8dc5b24b637eb124b8cff8",
   "size_mb": 8370,
   "ram_peak_mb": 7387,
   "ram_mb": 7387
  },
  {
   "model": "winnow:12b",
   "questions": "triage",
   "load_ms": 32353.6,
   "first_request_ms": 43494.2,
   "answered_by": "winnow:12b",
   "p50_ms": 11242.9,
   "p90_ms": 11338.9,
   "min_ms": 11043.2,
   "eval_p50_ms": 11241.7,
   "input_tokens": 711,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "8d72fee0e5f92a625cdb85ae4f6abdc024c9e6f124a0ca67e64522c3f24c47ce",
   "size_mb": 12083,
   "ram_peak_mb": 14965,
   "ram_mb": 14964
  },
  {
   "model": "cygnet:12b",
   "questions": "triage",
   "load_ms": 41616.9,
   "first_request_ms": 62218.2,
   "answered_by": "cygnet:12b",
   "p50_ms": 20664.2,
   "p90_ms": 20975.7,
   "min_ms": 19415.1,
   "eval_p50_ms": 20663,
   "input_tokens": 746,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "81160d85e13acfda209f78ba38a3e1d7ef0c80727820b022bbd9c878476e93cc",
   "size_mb": 12083,
   "ram_peak_mb": 17648,
   "ram_mb": 17648
  },
  {
   "model": "clm:8b",
   "questions": "triage",
   "load_ms": 36123.7,
   "first_request_ms": 59034.7,
   "answered_by": "clm:8b",
   "p50_ms": 5230.2,
   "p90_ms": 5548.5,
   "min_ms": 5041.3,
   "eval_p50_ms": 5229.3,
   "input_tokens": 357,
   "device": "cpu",
   "precision": "F32",
   "digest": "3f89a66972f1398210d2f41b94667665132aa51c2d313237c3fc70a49c275cb5",
   "size_mb": 15712,
   "ram_peak_mb": 14808,
   "ram_mb": 14808
  },
  {
   "model": "jeb:4b",
   "questions": "triage",
   "load_ms": 11129.7,
   "first_request_ms": 16528.1,
   "answered_by": "jeb:4b",
   "p50_ms": 5333.8,
   "p90_ms": 5435.4,
   "min_ms": 4643,
   "eval_p50_ms": 5332.5,
   "input_tokens": 569,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "4aba27707f6f4ba6f116710a0d05743220e5a5819b8826a9bd32385d0d9df2ff",
   "size_mb": 4397,
   "ram_peak_mb": 4595,
   "ram_mb": 4595
  },
  {
   "model": "jeb:9b",
   "questions": "triage",
   "load_ms": 21571.4,
   "first_request_ms": 31388.3,
   "answered_by": "jeb:9b",
   "p50_ms": 9417.9,
   "p90_ms": 9711,
   "min_ms": 8947.6,
   "eval_p50_ms": 9416.7,
   "input_tokens": 574,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "dfe0411a8d060f1fa64ff2539b1e4b44a061c11cce4d5d96a82a7077b01d8b9e",
   "size_mb": 9333,
   "ram_peak_mb": 9434,
   "ram_mb": 9415
  },
  {
   "model": "clef:flash",
   "questions": "triage",
   "load_ms": 35898.7,
   "first_request_ms": 53192.6,
   "answered_by": "clef:flash",
   "p50_ms": 12477.2,
   "p90_ms": 12698.5,
   "min_ms": 12339,
   "eval_p50_ms": 12476.2,
   "input_tokens": 632,
   "device": "cpu",
   "precision": "F32",
   "digest": "acda6b87d74dc55e04916220fcfc643ca2d409ca22ec5e9d18bf53b91604845d",
   "size_mb": 18211,
   "ram_peak_mb": 16138,
   "ram_mb": 16138
  },
  {
   "model": "jeeves:9b",
   "questions": "triage",
   "load_ms": 34872.7,
   "first_request_ms": 55962.3,
   "answered_by": "jeeves:9b",
   "p50_ms": 20474.3,
   "p90_ms": 20558,
   "min_ms": 20305.7,
   "eval_p50_ms": 20473.2,
   "input_tokens": 503,
   "device": "cpu",
   "precision": "F32",
   "digest": "0f4a1ff8396c0718c432603d96c0a8eb975a0d3bab83bd374bd35e36dfc5512d",
   "size_mb": 17116,
   "ram_peak_mb": 16912,
   "ram_mb": 16912
  },
  {
   "model": "nimble:9b",
   "questions": "triage",
   "load_ms": 41091.2,
   "first_request_ms": 105297.9,
   "answered_by": "nimble:9b",
   "p50_ms": 62005.5,
   "p90_ms": 62148.5,
   "min_ms": 61876.3,
   "eval_p50_ms": 62004.4,
   "input_tokens": 2983,
   "device": "cpu",
   "precision": "F32",
   "digest": "fb7e7f742e34636e853f91c467b9463e3155dbb08cabc5c49906d52a5750487d",
   "size_mb": 18610,
   "ram_peak_mb": 19041,
   "ram_mb": 19041
  },
  {
   "model": "jeb:27b",
   "questions": "triage",
   "load_ms": 55205.6,
   "first_request_ms": 76520.6,
   "answered_by": "jeb:27b",
   "p50_ms": 20990.3,
   "p90_ms": 21377.5,
   "min_ms": 20727.2,
   "eval_p50_ms": 20989.1,
   "input_tokens": 579,
   "device": "cpu",
   "precision": "Q4_K - MEDIUM",
   "digest": "da59bd8b0f90123b7ca0de07f2a325aa860b8ab97814ba36226f11463186f561",
   "size_mb": 16032,
   "ram_peak_mb": 27189,
   "ram_mb": 27189
  }
 ],
 "edition": "2026.10"
}
