{
 "schema": 1,
 "suite": "latency-triage",
 "date": "2026-10-24T12:00:17Z",
 "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": "gliclass:v4-edge",
   "questions": "triage",
   "load_ms": 8135.2,
   "first_request_ms": 8165.6,
   "answered_by": "gliclass:v4-edge",
   "p50_ms": 16.3,
   "p90_ms": 16.9,
   "min_ms": 14.5,
   "eval_p50_ms": 15.3,
   "input_tokens": 273,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "3b1849b2b5ec38109cbf8772715bccaf8f546ab48f6ecefdee41f4d6af1c5b39",
   "size_mb": 854,
   "ram_peak_mb": 4104,
   "ram_mb": 3610,
   "vram_mb": 2677,
   "gpu_idle_mb": 242
  },
  {
   "model": "decima:base-clean",
   "questions": "triage",
   "load_ms": 3364.5,
   "first_request_ms": 3392.3,
   "answered_by": "decima:base-clean",
   "p50_ms": 26.4,
   "p90_ms": 33.1,
   "min_ms": 25.2,
   "eval_p50_ms": 25.4,
   "input_tokens": 458,
   "device": "cuda:0",
   "precision": "F32",
   "digest": "b3d4c59e951d30fef560f8a2ed6d2b2c1355cfae84b4b538dbefdea2fbcf289c",
   "size_mb": 1263,
   "ram_peak_mb": 1824,
   "ram_mb": 1094,
   "vram_mb": 2933,
   "gpu_idle_mb": 242
  },
  {
   "model": "winnow:e2b",
   "questions": "triage",
   "load_ms": 3862.4,
   "first_request_ms": 3974.5,
   "answered_by": "winnow:e2b",
   "p50_ms": 103.6,
   "p90_ms": 105.9,
   "min_ms": 92.2,
   "eval_p50_ms": 102.6,
   "input_tokens": 691,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "75b519b33318ec29f6a734f514f5e7636614e78a07e552cbeb7f98d6672ca808",
   "size_mb": 4725,
   "ram_peak_mb": 5168,
   "ram_mb": 3329,
   "vram_mb": 3609,
   "gpu_idle_mb": 242
  },
  {
   "model": "jeb:9b",
   "questions": "triage",
   "load_ms": 9591.9,
   "first_request_ms": 9943.2,
   "answered_by": "jeb:9b",
   "p50_ms": 111.2,
   "p90_ms": 119.3,
   "min_ms": 108.5,
   "eval_p50_ms": 110.1,
   "input_tokens": 569,
   "device": "cuda:0",
   "precision": "Q8_0",
   "digest": "737d66d5ad2aaf17e4271257a8b7853a6613967d6953bda3c8167fb8e6d5270c",
   "size_mb": 9333,
   "ram_peak_mb": 9744,
   "ram_mb": 1532,
   "vram_mb": 9315,
   "gpu_idle_mb": 242
  },
  {
   "model": "jeb:27b",
   "questions": "triage",
   "load_ms": 9170.5,
   "first_request_ms": 9510,
   "answered_by": "jeb:27b",
   "p50_ms": 263.7,
   "p90_ms": 269.3,
   "min_ms": 253.1,
   "eval_p50_ms": 262.7,
   "input_tokens": 569,
   "device": "cuda:0",
   "precision": "Q4_K - MEDIUM",
   "digest": "978465d647d1de010ea24370f135b3b2552020208612163c5907abe92de465de",
   "size_mb": 16032,
   "ram_peak_mb": 16443,
   "ram_mb": 1205,
   "vram_mb": 16591,
   "gpu_idle_mb": 242
  }
 ],
 "edition": "2026.10"
}