{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-20T09:19:49Z",
 "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": "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 0.14.0. Release candidate (rc/0.15.0), measured before the release; one model loaded at a time.",
 "results": [
  {
   "model": "largha:150m",
   "questions": "triage",
   "load_ms": 3804.3,
   "first_request_ms": 3843.3,
   "answered_by": "largha:150m",
   "p50_ms": 17.8,
   "p90_ms": 23.6,
   "min_ms": 14.3,
   "eval_p50_ms": 16.7,
   "input_tokens": 351,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "d5727855d1b4f8eb43180058b038c34560971c1833098bbe7d9140c27a142c5d",
   "size_mb": 313,
   "ram_peak_mb": 991,
   "ram_mb": 991,
   "vram_mb": 1011,
   "gpu_idle_mb": 250
  },
  {
   "model": "decima:agent",
   "questions": "triage",
   "load_ms": 7049.4,
   "first_request_ms": 7088.7,
   "answered_by": "decima:agent",
   "p50_ms": 25.3,
   "p90_ms": 31.5,
   "min_ms": 24.3,
   "eval_p50_ms": 24.3,
   "input_tokens": 458,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "e14e614ce4ab1ae2e7a0f1203b91e4d38fb645d100202726abe59f78e2c0077b",
   "size_mb": 1263,
   "ram_peak_mb": 1822,
   "ram_mb": 1124,
   "vram_mb": 2805,
   "gpu_idle_mb": 250
  },
  {
   "model": "decima:base",
   "questions": "triage",
   "load_ms": 3383.8,
   "first_request_ms": 3411.8,
   "answered_by": "decima:base",
   "p50_ms": 26.7,
   "p90_ms": 31.5,
   "min_ms": 24.9,
   "eval_p50_ms": 25.6,
   "input_tokens": 458,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "04092c1005aefaa65400483b23a264afe11ac617b644c081b3e8dc2badba5bd9",
   "size_mb": 1263,
   "ram_peak_mb": 1824,
   "ram_mb": 1165,
   "vram_mb": 2805,
   "gpu_idle_mb": 250
  },
  {
   "model": "decima:guard",
   "questions": "triage",
   "load_ms": 3277.8,
   "first_request_ms": 3305.1,
   "answered_by": "decima:guard",
   "p50_ms": 26.3,
   "p90_ms": 32.9,
   "min_ms": 25.1,
   "eval_p50_ms": 25.4,
   "input_tokens": 463,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "6365b94a39ba2a6c385fd3e31746bcae24b863d07eb4b2f3b0335c812e8e76cf",
   "size_mb": 1263,
   "ram_peak_mb": 1824,
   "ram_mb": 1127,
   "vram_mb": 2745,
   "gpu_idle_mb": 250
  },
  {
   "model": "decima:guard-small",
   "questions": "triage",
   "load_ms": 2643.2,
   "first_request_ms": 2676,
   "answered_by": "decima:guard-small",
   "p50_ms": 26.6,
   "p90_ms": 32.7,
   "min_ms": 24.5,
   "eval_p50_ms": 25.6,
   "input_tokens": 453,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "1285576f6cd12494c9ad22cfb754743c1e8d0ba7eea60e7c4871eb1f06bf96df",
   "size_mb": 592,
   "ram_peak_mb": 1109,
   "ram_mb": 945,
   "vram_mb": 1269,
   "gpu_idle_mb": 250
  },
  {
   "model": "decima:small",
   "questions": "triage",
   "load_ms": 2128.7,
   "first_request_ms": 2144.7,
   "answered_by": "decima:small",
   "p50_ms": 14.2,
   "p90_ms": 21.6,
   "min_ms": 13.5,
   "eval_p50_ms": 13.3,
   "input_tokens": 567,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "ae8ba9b80ee181b6c10f8199870e5fca984f187d8ffa7880ab9cb728957a93a3",
   "size_mb": 486,
   "ram_peak_mb": 1218,
   "ram_mb": 1135,
   "vram_mb": 1269,
   "gpu_idle_mb": 250
  }
 ],
 "edition": "2026.10"
}