{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-24T11:35:43Z",
 "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": "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 0.17.0.",
 "results": [
  {
   "model": "d3:flash",
   "questions": "triage",
   "load_ms": 39115.3,
   "first_request_ms": 39514.4,
   "answered_by": "d3:flash",
   "p50_ms": 317.1,
   "p90_ms": 320.3,
   "min_ms": 310.5,
   "eval_p50_ms": 316,
   "input_tokens": 656,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "0f5b2d171ede0443170352adff119afc54283be96d50bd6c6cf8e91380ea41a7",
   "size_mb": 16042,
   "ram_peak_mb": 2979,
   "ram_mb": 1562,
   "vram_mb": 18397,
   "gpu_idle_mb": 242
  },
  {
   "model": "d3:edge",
   "questions": "triage",
   "load_ms": 6875.7,
   "first_request_ms": 6917.9,
   "answered_by": "d3:edge",
   "p50_ms": 27.7,
   "p90_ms": 37.3,
   "min_ms": 26.2,
   "eval_p50_ms": 26.6,
   "input_tokens": 656,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "88b3e621e7c659b907938f2a79564734f1f341d0eaace91f9389cdc50555e462",
   "size_mb": 1158,
   "ram_peak_mb": 3121,
   "ram_mb": 3104,
   "vram_mb": 2657,
   "gpu_idle_mb": 242
  },
  {
   "model": "d3:lite",
   "questions": "triage",
   "load_ms": 9425.1,
   "first_request_ms": 9548.8,
   "answered_by": "d3:lite",
   "p50_ms": 65.2,
   "p90_ms": 68.4,
   "min_ms": 55.5,
   "eval_p50_ms": 64.1,
   "input_tokens": 656,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "1f9c3e2dd6c2880348e1000ed93088a98b49ee5f298ca65eec6c49d433ffe53e",
   "size_mb": 1655,
   "ram_peak_mb": 5542,
   "ram_mb": 5542,
   "vram_mb": 4539,
   "gpu_idle_mb": 242
  },
  {
   "model": "d3:nano",
   "questions": "triage",
   "load_ms": 30201,
   "first_request_ms": 30313.7,
   "answered_by": "d3:nano",
   "p50_ms": 88.9,
   "p90_ms": 92.2,
   "min_ms": 78.7,
   "eval_p50_ms": 87.9,
   "input_tokens": 656,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "137bd4f1edd15d843102431d9b4ed066e581549fc3a5e61f72e6d09a28519ecf",
   "size_mb": 4251,
   "ram_peak_mb": 11704,
   "ram_mb": 1219,
   "vram_mb": 8363,
   "gpu_idle_mb": 242
  },
  {
   "model": "d3:mini",
   "questions": "triage",
   "load_ms": 17749.1,
   "first_request_ms": 18165.8,
   "answered_by": "d3:mini",
   "p50_ms": 201.8,
   "p90_ms": 204.4,
   "min_ms": 194.1,
   "eval_p50_ms": 200.7,
   "input_tokens": 656,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "5c148d7803097eb575b7b90308496c3f8ec61890a06d5515098c4279aa7b7e39",
   "size_mb": 8691,
   "ram_peak_mb": 2021,
   "ram_mb": 1324,
   "vram_mb": 11097,
   "gpu_idle_mb": 242
  }
 ],
 "edition": "2026.10"
}