{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-24T11:29:20Z",
 "ollaya": "0.17.0",
 "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": "Ollaya 0.17.0.",
 "results": [
  {
   "model": "d3:edge",
   "questions": "triage",
   "load_ms": 6375.4,
   "first_request_ms": 7659.7,
   "answered_by": "d3:edge",
   "p50_ms": 1247.8,
   "p90_ms": 1258.5,
   "min_ms": 1243.1,
   "eval_p50_ms": 1246.8,
   "input_tokens": 656,
   "device": "cpu",
   "precision": "F32",
   "digest": "88b3e621e7c659b907938f2a79564734f1f341d0eaace91f9389cdc50555e462",
   "size_mb": 1158,
   "ram_peak_mb": 2310,
   "ram_mb": 2051
  },
  {
   "model": "d3:lite",
   "questions": "triage",
   "load_ms": 12699.1,
   "first_request_ms": 15791.3,
   "answered_by": "d3:lite",
   "p50_ms": 2951.2,
   "p90_ms": 2981.1,
   "min_ms": 2938.1,
   "eval_p50_ms": 2950.2,
   "input_tokens": 656,
   "device": "cpu",
   "precision": "F32",
   "digest": "1f9c3e2dd6c2880348e1000ed93088a98b49ee5f298ca65eec6c49d433ffe53e",
   "size_mb": 1655,
   "ram_peak_mb": 4705,
   "ram_mb": 3086
  },
  {
   "model": "d3:nano",
   "questions": "triage",
   "load_ms": 33544.7,
   "first_request_ms": 37660.7,
   "answered_by": "d3:nano",
   "p50_ms": 3934.1,
   "p90_ms": 3958.5,
   "min_ms": 3887.3,
   "eval_p50_ms": 3933.1,
   "input_tokens": 656,
   "device": "cpu",
   "precision": "F32",
   "digest": "137bd4f1edd15d843102431d9b4ed066e581549fc3a5e61f72e6d09a28519ecf",
   "size_mb": 4251,
   "ram_peak_mb": 11386,
   "ram_mb": 6326
  },
  {
   "model": "d3:mini",
   "questions": "triage",
   "load_ms": 21168.1,
   "first_request_ms": 31090.6,
   "answered_by": "d3:mini",
   "p50_ms": 9586.1,
   "p90_ms": 9636.7,
   "min_ms": 9520.9,
   "eval_p50_ms": 9585.1,
   "input_tokens": 656,
   "device": "cpu",
   "precision": "F32",
   "digest": "5c148d7803097eb575b7b90308496c3f8ec61890a06d5515098c4279aa7b7e39",
   "size_mb": 8691,
   "ram_peak_mb": 9382,
   "ram_mb": 9382
  },
  {
   "model": "d3:flash",
   "questions": "triage",
   "load_ms": 19650.4,
   "first_request_ms": 35650.8,
   "answered_by": "d3:flash",
   "p50_ms": 14844,
   "p90_ms": 14893.4,
   "min_ms": 14719.1,
   "eval_p50_ms": 14842.9,
   "input_tokens": 656,
   "device": "cpu",
   "precision": "F32",
   "digest": "0f5b2d171ede0443170352adff119afc54283be96d50bd6c6cf8e91380ea41a7",
   "size_mb": 16042,
   "ram_peak_mb": 16383,
   "ram_mb": 16383
  }
 ],
 "edition": "2026.10"
}