{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-24T12:09:34Z",
 "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": "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 0.17.0.",
 "results": [
  {
   "model": "gliclass:v4-edge",
   "questions": "triage",
   "load_ms": 9040.4,
   "first_request_ms": 10919.6,
   "answered_by": "gliclass:v4-edge",
   "p50_ms": 1847.9,
   "p90_ms": 1912.7,
   "min_ms": 1650.8,
   "eval_p50_ms": 1846.9,
   "input_tokens": 273,
   "device": "cpu",
   "precision": "F32",
   "digest": "3b1849b2b5ec38109cbf8772715bccaf8f546ab48f6ecefdee41f4d6af1c5b39",
   "size_mb": 854,
   "ram_peak_mb": 3842,
   "ram_mb": 3842
  },
  {
   "model": "decima:base-clean",
   "questions": "triage",
   "load_ms": 4435.1,
   "first_request_ms": 5707.5,
   "answered_by": "decima:base-clean",
   "p50_ms": 1198.5,
   "p90_ms": 1272.2,
   "min_ms": 1161.7,
   "eval_p50_ms": 1197.4,
   "input_tokens": 458,
   "device": "cpu",
   "precision": "F32",
   "digest": "b3d4c59e951d30fef560f8a2ed6d2b2c1355cfae84b4b538dbefdea2fbcf289c",
   "size_mb": 1263,
   "ram_peak_mb": 1124,
   "ram_mb": 1124
  },
  {
   "model": "winnow:e2b",
   "questions": "triage",
   "load_ms": 10533.4,
   "first_request_ms": 12604.4,
   "answered_by": "winnow:e2b",
   "p50_ms": 2044.9,
   "p90_ms": 2057.3,
   "min_ms": 2001.3,
   "eval_p50_ms": 2043.8,
   "input_tokens": 691,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "75b519b33318ec29f6a734f514f5e7636614e78a07e552cbeb7f98d6672ca808",
   "size_mb": 4725,
   "ram_peak_mb": 5031,
   "ram_mb": 5031
  },
  {
   "model": "jeb:9b",
   "questions": "triage",
   "load_ms": 21811,
   "first_request_ms": 31203.3,
   "answered_by": "jeb:9b",
   "p50_ms": 9429.9,
   "p90_ms": 9786.4,
   "min_ms": 9059,
   "eval_p50_ms": 9428.8,
   "input_tokens": 569,
   "device": "cpu",
   "precision": "Q8_0",
   "digest": "737d66d5ad2aaf17e4271257a8b7853a6613967d6953bda3c8167fb8e6d5270c",
   "size_mb": 9333,
   "ram_peak_mb": 9434,
   "ram_mb": 9415
  },
  {
   "model": "jeb:27b",
   "questions": "triage",
   "load_ms": 27252.8,
   "first_request_ms": 48137.2,
   "answered_by": "jeb:27b",
   "p50_ms": 20667.6,
   "p90_ms": 21065.5,
   "min_ms": 20339.1,
   "eval_p50_ms": 20666.4,
   "input_tokens": 569,
   "device": "cpu",
   "precision": "Q4_K - MEDIUM",
   "digest": "978465d647d1de010ea24370f135b3b2552020208612163c5907abe92de465de",
   "size_mb": 16032,
   "ram_peak_mb": 27186,
   "ram_mb": 27186
  }
 ],
 "edition": "2026.10"
}