{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-20T09:27:07Z",
 "ollaya": "0.14.0",
 "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 0.14.0. Release candidate (rc/0.15.0), measured before the release; one model loaded at a time.",
 "results": [
  {
   "model": "largha:150m",
   "questions": "triage",
   "load_ms": 2268.2,
   "first_request_ms": 2649.1,
   "answered_by": "largha:150m",
   "p50_ms": 339.1,
   "p90_ms": 343.7,
   "min_ms": 335.3,
   "eval_p50_ms": 338,
   "input_tokens": 351,
   "device": "cpu",
   "precision": "F32",
   "digest": "d5727855d1b4f8eb43180058b038c34560971c1833098bbe7d9140c27a142c5d",
   "size_mb": 313,
   "ram_peak_mb": 992,
   "ram_mb": 992
  },
  {
   "model": "decima:agent",
   "questions": "triage",
   "load_ms": 3186.7,
   "first_request_ms": 4479.8,
   "answered_by": "decima:agent",
   "p50_ms": 1239.5,
   "p90_ms": 1327.7,
   "min_ms": 1172.8,
   "eval_p50_ms": 1238.4,
   "input_tokens": 458,
   "device": "cpu",
   "precision": "F32",
   "digest": "e14e614ce4ab1ae2e7a0f1203b91e4d38fb645d100202726abe59f78e2c0077b",
   "size_mb": 1263,
   "ram_peak_mb": 1133,
   "ram_mb": 1133
  },
  {
   "model": "decima:base",
   "questions": "triage",
   "load_ms": 2942,
   "first_request_ms": 4099.6,
   "answered_by": "decima:base",
   "p50_ms": 1212.5,
   "p90_ms": 1276.7,
   "min_ms": 1190.1,
   "eval_p50_ms": 1211.5,
   "input_tokens": 458,
   "device": "cpu",
   "precision": "F32",
   "digest": "04092c1005aefaa65400483b23a264afe11ac617b644c081b3e8dc2badba5bd9",
   "size_mb": 1263,
   "ram_peak_mb": 1128,
   "ram_mb": 1128
  },
  {
   "model": "decima:guard",
   "questions": "triage",
   "load_ms": 3156.5,
   "first_request_ms": 4376.5,
   "answered_by": "decima:guard",
   "p50_ms": 1244.9,
   "p90_ms": 1284.3,
   "min_ms": 1185.9,
   "eval_p50_ms": 1243.9,
   "input_tokens": 458,
   "device": "cpu",
   "precision": "F32",
   "digest": "6365b94a39ba2a6c385fd3e31746bcae24b863d07eb4b2f3b0335c812e8e76cf",
   "size_mb": 1263,
   "ram_peak_mb": 1126,
   "ram_mb": 1126
  },
  {
   "model": "decima:guard-small",
   "questions": "triage",
   "load_ms": 2087.7,
   "first_request_ms": 2899.4,
   "answered_by": "decima:guard-small",
   "p50_ms": 720.6,
   "p90_ms": 738.7,
   "min_ms": 701.6,
   "eval_p50_ms": 719.6,
   "input_tokens": 478,
   "device": "cpu",
   "precision": "F32",
   "digest": "1285576f6cd12494c9ad22cfb754743c1e8d0ba7eea60e7c4871eb1f06bf96df",
   "size_mb": 592,
   "ram_peak_mb": 529,
   "ram_mb": 529
  },
  {
   "model": "decima:small",
   "questions": "triage",
   "load_ms": 1295,
   "first_request_ms": 1952.4,
   "answered_by": "decima:small",
   "p50_ms": 591.4,
   "p90_ms": 609.4,
   "min_ms": 574.5,
   "eval_p50_ms": 590.4,
   "input_tokens": 567,
   "device": "cpu",
   "precision": "F32",
   "digest": "ae8ba9b80ee181b6c10f8199870e5fca984f187d8ffa7880ab9cb728957a93a3",
   "size_mb": 486,
   "ram_peak_mb": 586,
   "ram_mb": 586
  }
 ],
 "edition": "2026.10"
}