{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-25T18:07:55Z",
 "ollaya": "0.18.0",
 "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": "Ollaya 0.18.0.",
 "results": [
  {
   "model": "d1-omni:600m",
   "questions": "triage",
   "load_ms": 7899.5,
   "first_request_ms": 7918.6,
   "answered_by": "d1-omni:600m",
   "p50_ms": 16.6,
   "p90_ms": 17.4,
   "min_ms": 15.5,
   "eval_p50_ms": 15.5,
   "input_tokens": 374,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "53dfa4be304b8ab233617f432ed655050d6d459a48f89dae94a574f75414f139",
   "size_mb": 2248,
   "ram_peak_mb": 1971,
   "ram_mb": 1789,
   "vram_mb": 2681,
   "gpu_idle_mb": 195
  },
  {
   "model": "gliclass-v4:2b",
   "questions": "triage",
   "load_ms": 50871.2,
   "first_request_ms": 51061.8,
   "answered_by": "gliclass-v4:2b",
   "p50_ms": 82.2,
   "p90_ms": 83.7,
   "min_ms": 66.7,
   "eval_p50_ms": 81.2,
   "input_tokens": 270,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "e2575bb26e1362f95d78049e2988ba36973ebb95f308b7d3c94aa277dd05a870",
   "size_mb": 4545,
   "ram_peak_mb": 12927,
   "ram_mb": 3778,
   "vram_mb": 9127,
   "gpu_idle_mb": 195
  },
  {
   "model": "gliclass-v4:9b",
   "questions": "triage",
   "load_ms": 44493.4,
   "first_request_ms": 45038.7,
   "answered_by": "gliclass-v4:9b",
   "p50_ms": 263.4,
   "p90_ms": 267.4,
   "min_ms": 250.8,
   "eval_p50_ms": 261.8,
   "input_tokens": 270,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "0e5bbb07b002e136f66bd57e8dd5bf42613df06dfbb55c80ebd016019e737529",
   "size_mb": 17170,
   "ram_peak_mb": 3000,
   "ram_mb": 1717,
   "vram_mb": 21493,
   "gpu_idle_mb": 195
  },
  {
   "model": "strands-decider:e2b",
   "questions": "triage",
   "load_ms": 30529.7,
   "first_request_ms": 30837.1,
   "answered_by": "strands-decider:e2b",
   "p50_ms": 116.2,
   "p90_ms": 130.7,
   "min_ms": 115.6,
   "eval_p50_ms": 115.1,
   "input_tokens": 570,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "76e87da292235d224abe1f24f19596c5280c9d9eabe4e0ff3278697de088d438",
   "size_mb": 10099,
   "ram_peak_mb": 5286,
   "ram_mb": 1207,
   "vram_mb": 10323,
   "gpu_idle_mb": 195
  },
  {
   "model": "strands-decider:e4b",
   "questions": "triage",
   "load_ms": 48625,
   "first_request_ms": 48860.9,
   "answered_by": "strands-decider:e4b",
   "p50_ms": 212.9,
   "p90_ms": 226.3,
   "min_ms": 211.2,
   "eval_p50_ms": 211.9,
   "input_tokens": 570,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "b923eec149f21f2b2cc2a294c9b1da22c52cbbdb72b0f6d044e63054765809c3",
   "size_mb": 15710,
   "ram_peak_mb": 6360,
   "ram_mb": 1435,
   "vram_mb": 15813,
   "gpu_idle_mb": 195
  },
  {
   "model": "decider:2b",
   "questions": "triage",
   "load_ms": 34725.2,
   "first_request_ms": 34821.3,
   "answered_by": "decider:2b",
   "p50_ms": 84.1,
   "p90_ms": 91.7,
   "min_ms": 75.9,
   "eval_p50_ms": 82.9,
   "input_tokens": 552,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "6b81ac65a337f2888699e594e1f325aefe7e9718ce0fba8d6da4a3492aba1982",
   "size_mb": 3616,
   "ram_peak_mb": 11576,
   "ram_mb": 3059,
   "vram_mb": 8463,
   "gpu_idle_mb": 195
  },
  {
   "model": "decider:4b",
   "questions": "triage",
   "load_ms": 22682.7,
   "first_request_ms": 23185.8,
   "answered_by": "decider:4b",
   "p50_ms": 196.5,
   "p90_ms": 208.1,
   "min_ms": 193.3,
   "eval_p50_ms": 195.5,
   "input_tokens": 552,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "b4e1276b55eda06fad7af9536a996fc132a1229bb866755cd6f07778dbe98963",
   "size_mb": 8052,
   "ram_peak_mb": 1969,
   "ram_mb": 1337,
   "vram_mb": 11279,
   "gpu_idle_mb": 195
  },
  {
   "model": "sura:0.8b",
   "questions": "triage",
   "load_ms": 13109.8,
   "first_request_ms": 13197.4,
   "answered_by": "sura:0.8b",
   "p50_ms": 69.3,
   "p90_ms": 73.6,
   "min_ms": 56.5,
   "eval_p50_ms": 68.2,
   "input_tokens": 320,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "473fa350466df82396f2a19152bdc15abaa79a6e57fc99f867c3e0dd280ae18a",
   "size_mb": 1654,
   "ram_peak_mb": 5574,
   "ram_mb": 5544,
   "vram_mb": 4293,
   "gpu_idle_mb": 195
  }
 ],
 "edition": "2026.10"
}