{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-25T18:00:31Z",
 "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": "decision2:kai",
   "questions": "triage",
   "load_ms": 11250.2,
   "first_request_ms": 11332.9,
   "answered_by": "decision2:kai",
   "p50_ms": 30.8,
   "p90_ms": 44,
   "min_ms": 30.4,
   "eval_p50_ms": 29.8,
   "input_tokens": 587,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "c1f010e05fb8fab53826338931bbd5b5cb7b10ad8410fdf5bf1c344bd6e097f5",
   "size_mb": 1454,
   "ram_peak_mb": 4957,
   "ram_mb": 4670,
   "vram_mb": 3701,
   "gpu_idle_mb": 195
  },
  {
   "model": "decision2:eos",
   "questions": "triage",
   "load_ms": 18732.1,
   "first_request_ms": 18862.5,
   "answered_by": "decision2:eos",
   "p50_ms": 68.5,
   "p90_ms": 76.6,
   "min_ms": 60.7,
   "eval_p50_ms": 67.4,
   "input_tokens": 644,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "4c8c10ac75b300337e9687b249e879a19287fc4481f23e5c5e98020eb650733b",
   "size_mb": 1952,
   "ram_peak_mb": 6090,
   "ram_mb": 5632,
   "vram_mb": 5909,
   "gpu_idle_mb": 195
  },
  {
   "model": "decision2:sol",
   "questions": "triage",
   "load_ms": 44409.1,
   "first_request_ms": 44515.7,
   "answered_by": "decision2:sol",
   "p50_ms": 89.4,
   "p90_ms": 103.7,
   "min_ms": 88.3,
   "eval_p50_ms": 88.3,
   "input_tokens": 644,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "85d3edf9e8bcf7cb39f0907c504781fd2f57a6c0725a24aa4f202dbe103f28a9",
   "size_mb": 4596,
   "ram_peak_mb": 11719,
   "ram_mb": 1483,
   "vram_mb": 11027,
   "gpu_idle_mb": 195
  },
  {
   "model": "decision2:nox",
   "questions": "triage",
   "load_ms": 26296.3,
   "first_request_ms": 27107.6,
   "answered_by": "decision2:nox",
   "p50_ms": 250.5,
   "p90_ms": 259.2,
   "min_ms": 239.1,
   "eval_p50_ms": 249,
   "input_tokens": 644,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "d81d2a4fe5cad9b1e0e6a8c44d9108732a129328a9e595c935c4b6083b5e9133",
   "size_mb": 9277,
   "ram_peak_mb": 3250,
   "ram_mb": 1365,
   "vram_mb": 12387,
   "gpu_idle_mb": 195
  },
  {
   "model": "decision2:lux",
   "questions": "triage",
   "load_ms": 54811,
   "first_request_ms": 55731.4,
   "answered_by": "decision2:lux",
   "p50_ms": 364.3,
   "p90_ms": 373,
   "min_ms": 352.8,
   "eval_p50_ms": 363.3,
   "input_tokens": 644,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "4148abeac890bfcf50bab4fdae636336468b21a9291f9a3f42ae526b17fd32a8",
   "size_mb": 17127,
   "ram_peak_mb": 4814,
   "ram_mb": 1622,
   "vram_mb": 20349,
   "gpu_idle_mb": 195
  }
 ],
 "edition": "2026.10"
}