{
  "cpu": {
    "run_id": "20260219T164800Z_opt5_full_v2",
    "created_utc": "2026-02-19T01:21:33.573737+00:00",
    "host": "d9aff3ced886",
    "device": "cpu",
    "batch": 64,
    "hidden": 2048,
    "iters": 40,
    "warmup": 10,
    "metadata_path": "results/artifacts/json/20260219T164800Z_opt5_full_v2__cpu.metadata.json",
    "adapters": [
      {
        "adapter": "xla",
        "display_name": "XLA",
        "mode": "native",
        "mean_ms": 24.6795,
        "p50_ms": 15.377,
        "p95_ms": 53.956,
        "throughput_tokens_per_sec": 5310958.0
      },
      {
        "adapter": "pyc",
        "display_name": "PyC CUDA",
        "mode": "native",
        "mean_ms": 27.3495,
        "p50_ms": 26.342,
        "p95_ms": 31.373,
        "throughput_tokens_per_sec": 4792486.88
      },
      {
        "adapter": "torch_compile",
        "display_name": "PyTorch Compile",
        "mode": "native",
        "mean_ms": 67.3705,
        "p50_ms": 91.9345,
        "p95_ms": 96.006,
        "throughput_tokens_per_sec": 1945540.46
      },
      {
        "adapter": "tvm",
        "display_name": "TVM",
        "mode": "native",
        "mean_ms": 70.1233,
        "p50_ms": 93.2592,
        "p95_ms": 94.4789,
        "throughput_tokens_per_sec": 1869164.22
      },
      {
        "adapter": "glow",
        "display_name": "Glow",
        "mode": "proxy",
        "mean_ms": 87.9144,
        "p50_ms": 97.8844,
        "p95_ms": 101.3468,
        "throughput_tokens_per_sec": 1490905.14
      },
      {
        "adapter": "torch_eager",
        "display_name": "PyTorch Eager",
        "mode": "native",
        "mean_ms": 92.913,
        "p50_ms": 98.275,
        "p95_ms": 104.1338,
        "throughput_tokens_per_sec": 1410696.45
      },
      {
        "adapter": "tensorrt",
        "display_name": "TensorRT",
        "mode": "proxy",
        "mean_ms": 102.7787,
        "p50_ms": 100.5979,
        "p95_ms": 104.0894,
        "throughput_tokens_per_sec": 1275283.89
      }
    ]
  },
  "gpu": {
    "run_id": "20260219T164800Z_opt5_full_v2",
    "created_utc": "2026-02-19T01:24:24.672845+00:00",
    "host": "d9aff3ced886",
    "device": "cuda",
    "batch": 64,
    "hidden": 2048,
    "iters": 40,
    "warmup": 10,
    "metadata_path": "results/artifacts/json/20260219T164800Z_opt5_full_v2__gpu.metadata.json",
    "adapters": [
      {
        "adapter": "torch_eager",
        "display_name": "PyTorch Eager",
        "mode": "native",
        "mean_ms": 0.1529,
        "p50_ms": 0.1529,
        "p95_ms": 0.154,
        "throughput_tokens_per_sec": 856966115.05
      },
      {
        "adapter": "glow",
        "display_name": "Glow",
        "mode": "proxy",
        "mean_ms": 0.1533,
        "p50_ms": 0.1528,
        "p95_ms": 0.1587,
        "throughput_tokens_per_sec": 854772290.62
      },
      {
        "adapter": "pyc",
        "display_name": "PyC CUDA",
        "mode": "native",
        "mean_ms": 0.1753,
        "p50_ms": 0.175,
        "p95_ms": 0.183,
        "throughput_tokens_per_sec": 747807730.71
      },
      {
        "adapter": "torch_compile",
        "display_name": "PyTorch Compile",
        "mode": "native",
        "mean_ms": 0.2057,
        "p50_ms": 0.205,
        "p95_ms": 0.2126,
        "throughput_tokens_per_sec": 637153918.18
      },
      {
        "adapter": "tensorrt",
        "display_name": "TensorRT",
        "mode": "native",
        "mean_ms": 0.2137,
        "p50_ms": 0.2071,
        "p95_ms": 0.2395,
        "throughput_tokens_per_sec": 613229097.73
      },
      {
        "adapter": "xla",
        "display_name": "XLA",
        "mode": "native",
        "mean_ms": 0.3194,
        "p50_ms": 0.3196,
        "p95_ms": 0.3261,
        "throughput_tokens_per_sec": 410325519.42
      },
      {
        "adapter": "tvm",
        "display_name": "TVM",
        "mode": "native",
        "mean_ms": 0.3385,
        "p50_ms": 0.3384,
        "p95_ms": 0.3397,
        "throughput_tokens_per_sec": 387262987.34
      }
    ]
  }
}