{
  "gpu": "NVIDIA GeForce RTX 3090",
  "cuda_version": "12.1",
  "pytorch_version": "2.1.0",
  "latency_median_ms": 4.503,
  "latency_p95_ms": 4.569,
  "latency_p99_ms": 4.614,
  "throughput": {
    "batch_1": {
      "throughput_per_sec": 235.69,
      "latency_ms": 4.243
    },
    "batch_8": {
      "throughput_per_sec": 1861.93,
      "latency_ms": 0.537
    },
    "batch_16": {
      "throughput_per_sec": 3571.04,
      "latency_ms": 0.28
    },
    "batch_32": {
      "throughput_per_sec": 4830.1,
      "latency_ms": 0.207
    },
    "batch_64": {
      "throughput_per_sec": 5271.67,
      "latency_ms": 0.19
    }
  },
  "timestamp": "2026-04-07 22:58:56",
  "provenance": {
    "kind": "synthetic_stub",
    "not_c4_onnx": true,
    "note": "TransformerEncoder stub latency proxy \u2014 not shipping C4 Dual/ONNX classifier throughput"
  }
}