{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-24T02:38:29Z",
 "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": "sura:2b",
   "questions": "triage",
   "load_ms": 25314.5,
   "first_request_ms": 25400.3,
   "answered_by": "sura:2b",
   "p50_ms": 62.6,
   "p90_ms": 68.3,
   "min_ms": 54.8,
   "eval_p50_ms": 61.7,
   "input_tokens": 320,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "33489bfce93fe2075f800bf55be89346f3083826eb14495a29a43b00b1cbfe42",
   "size_mb": 4249,
   "ram_peak_mb": 11708,
   "ram_mb": 1251,
   "vram_mb": 8126,
   "gpu_idle_mb": 17527
  }
 ],
 "edition": "2026.10"
}