{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-23T20:30:53Z",
 "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": "coba:1b",
   "questions": "triage",
   "load_ms": 13918.2,
   "first_request_ms": 13960.7,
   "answered_by": "coba:1b",
   "p50_ms": 20.1,
   "p90_ms": 24.5,
   "min_ms": 16.9,
   "eval_p50_ms": 19.1,
   "input_tokens": 349,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "7aaf4eaed0da754d7766714cd5b661c33f5f22476001f7e8fb0bd11f12eeacde",
   "size_mb": 4239,
   "ram_peak_mb": 4777,
   "ram_mb": 4499,
   "vram_mb": 3703,
   "gpu_idle_mb": 242
  },
  {
   "model": "coba:400m",
   "questions": "triage",
   "load_ms": 4436.5,
   "first_request_ms": 4471.7,
   "answered_by": "coba:400m",
   "p50_ms": 20.5,
   "p90_ms": 24.9,
   "min_ms": 17.2,
   "eval_p50_ms": 19.6,
   "input_tokens": 349,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "99c4f5b9e23aef32e9ea990a192a6990f5679a9b095ffde4dd7889839405b9e6",
   "size_mb": 1617,
   "ram_peak_mb": 2187,
   "ram_mb": 2115,
   "vram_mb": 1653,
   "gpu_idle_mb": 242
  },
  {
   "model": "coba:400m-ternary",
   "questions": "triage",
   "load_ms": 4254.1,
   "first_request_ms": 4292,
   "answered_by": "coba:400m-ternary",
   "p50_ms": 19.6,
   "p90_ms": 22.1,
   "min_ms": 17.7,
   "eval_p50_ms": 18.7,
   "input_tokens": 354,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "ec26808146ae191e76ec72691bd0178c2fb4804407436df68575af7fcdcee7f8",
   "size_mb": 1617,
   "ram_peak_mb": 2188,
   "ram_mb": 2116,
   "vram_mb": 1653,
   "gpu_idle_mb": 242
  },
  {
   "model": "coba:150m-ternary",
   "questions": "triage",
   "load_ms": 2200.2,
   "first_request_ms": 2228,
   "answered_by": "coba:150m-ternary",
   "p50_ms": 15.4,
   "p90_ms": 19,
   "min_ms": 12.5,
   "eval_p50_ms": 14.5,
   "input_tokens": 349,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "e1a3fad29866d5b4413179e84f3945d1b7fd79e5a9faa6d6f5e131ed8d18c020",
   "size_mb": 634,
   "ram_peak_mb": 1073,
   "ram_mb": 1073,
   "vram_mb": 1139,
   "gpu_idle_mb": 242
  },
  {
   "model": "coba:68m-ternary",
   "questions": "triage",
   "load_ms": 1699.3,
   "first_request_ms": 1723.7,
   "answered_by": "coba:68m-ternary",
   "p50_ms": 13.6,
   "p90_ms": 17.4,
   "min_ms": 11.5,
   "eval_p50_ms": 12.8,
   "input_tokens": 364,
   "device": "cuda:0",
   "precision": "F16",
   "digest": "93063c00de80a702a38b1acdd28747ae353fe49a569e3f0bdac61896c9b4d3fa",
   "size_mb": 293,
   "ram_peak_mb": 934,
   "ram_mb": 934,
   "vram_mb": 899,
   "gpu_idle_mb": 242
  },
  {
   "model": "sura:0.8b",
   "questions": "triage",
   "load_ms": 9448.2,
   "first_request_ms": 9561.4,
   "answered_by": "sura:0.8b",
   "p50_ms": 59.7,
   "p90_ms": 61.6,
   "min_ms": 54.6,
   "eval_p50_ms": 58.6,
   "input_tokens": 319,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "b5a8e8f48b6aeba1bb805c5941d992832458e8e5e868fd8f80302915ff063f7f",
   "size_mb": 1654,
   "ram_peak_mb": 5346,
   "ram_mb": 5346,
   "vram_mb": 4293,
   "gpu_idle_mb": 242
  }
 ],
 "edition": "2026.10"
}