{
  "schemaVersion": "0.1",
  "rigrouteLabsVersion": "0.1.0",
  "run": {
    "id": "20260927T170032Z-ef58c371",
    "startedAt": "2026-09-27T17:00:32.230102+00:00",
    "finishedAt": "2026-09-27T17:03:16.771328+00:00"
  },
  "workload": {
    "type": "llm-inference",
    "manifest": "standard-ai-v0.1",
    "modelClass": "14b-q4"
  },
  "hardware": {
    "host": {
      "os": "macOS",
      "osVersion": "26.6.2",
      "kernel": "Darwin 25.6.0",
      "arch": "arm64",
      "machineId": "1f14d1e6bce85456"
    },
    "cpu": {
      "manufacturer": "Apple",
      "model": "Apple M1 Pro",
      "physicalCores": 10,
      "logicalCores": 10
    },
    "memory": {
      "installedBytes": 34359738368,
      "availableAtStartBytes": 11895701504
    },
    "accelerators": [
      {
        "manufacturer": "Apple",
        "model": "Apple M1 Pro",
        "deviceId": "UNKNOWN",
        "memoryArchitecture": "unified-memory",
        "dedicatedVramBytes": null,
        "unifiedMemoryBytes": 34359738368,
        "driver": "UNKNOWN",
        "backend": "Metal"
      }
    ]
  },
  "environment": {
    "drivers": {},
    "runtime": {
      "llamaCppVersion": "0.5.0",
      "llamaBenchBuildNumber": "11146",
      "llamaCppCommit": "7fe450e19",
      "backendsCompiled": "BLAS,CPU,MTL"
    }
  },
  "benchmark": {
    "engine": "llama-bench",
    "engineVersion": "0.5.0 (build 11146, commit 7fe450e19)",
    "model": {
      "path": "<local-path-redacted>/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e",
      "filename": "sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e",
      "sizeBytes": 9276184896,
      "sha256": "a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e",
      "declaredIdentity": {
        "modelType": "qwen3 14B Q4_K - Medium",
        "paramCount": 14768307200,
        "source": "llama-bench JSON output (model_type / model_n_params)"
      },
      "manifestModelClass": "14b-q4"
    },
    "context": 8192,
    "promptTokensPerTest": 512,
    "generationTokensPerTest": 128,
    "repetitions": 5,
    "command": [
      "/opt/homebrew/bin/llama-bench",
      "-m",
      "<local-path-redacted>/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e",
      "-d",
      "8192",
      "-p",
      "512",
      "-n",
      "128",
      "-r",
      "5",
      "-o",
      "json",
      "-v"
    ],
    "deviceUsed": "BLAS,MTL",
    "backendUsed": "BLAS,MTL"
  },
  "measurements": {
    "promptTokensPerSecond": {
      "unit": "tok/s",
      "mean": 97.085284,
      "stddev": 0.094529,
      "samples": [
        97.1055,
        97.1638,
        96.9338,
        97.1608,
        97.0626
      ]
    },
    "generationTokensPerSecond": {
      "unit": "tok/s",
      "mean": 11.725878,
      "stddev": 0.003764,
      "samples": [
        11.7253,
        11.7247,
        11.7317,
        11.7264,
        11.7213
      ]
    },
    "durationSeconds": 80.948725916,
    "memory": {
      "acceleratorBytes": "UNAVAILABLE",
      "systemBytes": "UNAVAILABLE"
    },
    "power": "UNAVAILABLE",
    "temperature": "UNAVAILABLE"
  },
  "status": "PASS",
  "failure": null,
  "rawEvidence": {
    "command": [
      "/opt/homebrew/bin/llama-bench",
      "-m",
      "<local-path-redacted>/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e",
      "-d",
      "8192",
      "-p",
      "512",
      "-n",
      "128",
      "-r",
      "5",
      "-o",
      "json",
      "-v"
    ],
    "exitCode": 0,
    "stdoutPath": "stdout.log",
    "stderrPath": "stderr.log",
    "llamaBenchJson": [
      {
        "build_commit": "7fe450e19",
        "build_number": 11146,
        "cpu_info": "Accelerate, Apple M1 Pro",
        "gpu_info": "Apple M1 Pro",
        "backends": "BLAS,MTL",
        "model_filename": "<local-path-redacted>/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e",
        "model_type": "qwen3 14B Q4_K - Medium",
        "model_size": 9270225920,
        "model_n_params": 14768307200,
        "n_batch": 2048,
        "n_ubatch": 512,
        "n_threads": 8,
        "cpu_mask": "0x0",
        "cpu_strict": false,
        "poll": 50,
        "type_k": "f16",
        "type_v": "f16",
        "n_gpu_layers": -1,
        "n_cpu_moe": 0,
        "split_mode": "layer",
        "main_gpu": 0,
        "no_kv_offload": false,
        "flash_attn": -1,
        "devices": "auto",
        "tensor_split": "0.00",
        "tensor_buft_overrides": "none",
        "load_mode": "auto",
        "lazy_mode": "auto",
        "embeddings": false,
        "no_op_offload": 0,
        "no_host": false,
        "fit_target": 0,
        "fit_min_ctx": 0,
        "n_prompt": 512,
        "n_gen": 0,
        "n_depth": 8192,
        "test_time": "2026-09-27T17:00:39Z",
        "avg_ns": 5273717791,
        "stddev_ns": 5139432,
        "avg_ts": 97.085284,
        "stddev_ts": 0.094529,
        "samples_ns": [
          5272616291,
          5269454000,
          5281956584,
          5269614208,
          5274947875
        ],
        "samples_ts": [
          97.1055,
          97.1638,
          96.9338,
          97.1608,
          97.0626
        ]
      },
      {
        "build_commit": "7fe450e19",
        "build_number": 11146,
        "cpu_info": "Accelerate, Apple M1 Pro",
        "gpu_info": "Apple M1 Pro",
        "backends": "BLAS,MTL",
        "model_filename": "<local-path-redacted>/sha256-a8cc1361f3145dc01f6d77c6c82c9116b9ffe3c97b34716fe20418455876c40e",
        "model_type": "qwen3 14B Q4_K - Medium",
        "model_size": 9270225920,
        "model_n_params": 14768307200,
        "n_batch": 2048,
        "n_ubatch": 512,
        "n_threads": 8,
        "cpu_mask": "0x0",
        "cpu_strict": false,
        "poll": 50,
        "type_k": "f16",
        "type_v": "f16",
        "n_gpu_layers": -1,
        "n_cpu_moe": 0,
        "split_mode": "layer",
        "main_gpu": 0,
        "no_kv_offload": false,
        "flash_attn": -1,
        "devices": "auto",
        "tensor_split": "0.00",
        "tensor_buft_overrides": "none",
        "load_mode": "auto",
        "lazy_mode": "auto",
        "embeddings": false,
        "no_op_offload": 0,
        "no_host": false,
        "fit_target": 0,
        "fit_min_ctx": 0,
        "n_prompt": 0,
        "n_gen": 128,
        "n_depth": 8192,
        "test_time": "2026-09-27T17:02:21Z",
        "avg_ns": 10916027391,
        "stddev_ns": 3505918,
        "avg_ts": 11.725878,
        "stddev_ts": 0.003764,
        "samples_ns": [
          10916583250,
          10917102250,
          10910599291,
          10915584959,
          10920267208
        ],
        "samples_ts": [
          11.7253,
          11.7247,
          11.7317,
          11.7264,
          11.7213
        ]
      }
    ]
  }
}