{
  "generated_at": "2026-08-09T17:16:17.149860+00:00",
  "host": {
    "architecture": "arm64",
    "ollama_base_url": "http://127.0.0.1:11434",
    "processor": "Apple M5 Pro",
    "system": "Darwin"
  },
  "method": {
    "claim_boundary": "Local single-host benchmark. Quality is exact-match accuracy on the versioned task set; energy use and production workload performance are not measured.",
    "repeats": 2,
    "seed": 7,
    "tasks": 5,
    "temperature": 0,
    "warmup_runs_per_model": 1
  },
  "optimization": {
    "baseline_model": "qwen2.5:7b-instruct",
    "model_size_reduction_pct": 78.94,
    "optimized_model": "qwen2.5:1.5b-instruct-q4_K_M",
    "p95_latency_reduction_pct": 44.49,
    "quality_delta_points": -20.0,
    "resident_memory_reduction_pct": 75.47,
    "throughput_gain_pct": 184.68
  },
  "results": [
    {
      "cold_start_ms": 1037.329,
      "context_tokens": 32768,
      "load_duration_ms": 919.154,
      "median_tokens_per_second": 106.525,
      "model": "qwen2.5:7b-instruct",
      "model_size_bytes": 4683087332,
      "parameter_size": "7.6B",
      "passed_samples": 10,
      "quality_accuracy": 1.0,
      "quantization_level": "Q4_K_M",
      "resident_memory_bytes": 4740716952,
      "samples": [
        {
          "output_tokens": 4,
          "passed": true,
          "repeat": 1,
          "response": "518",
          "task_id": "arithmetic",
          "tokens_per_second": 72.709,
          "wall_latency_ms": 253.882
        },
        {
          "output_tokens": 7,
          "passed": true,
          "repeat": 1,
          "response": "{\"kind\":\"missing_owner\"}",
          "task_id": "governance",
          "tokens_per_second": 64.004,
          "wall_latency_ms": 286.334
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 1,
          "response": "reject",
          "task_id": "health_policy",
          "tokens_per_second": 107.77,
          "wall_latency_ms": 202.081
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 1,
          "response": "B",
          "task_id": "budget_route",
          "tokens_per_second": 108.719,
          "wall_latency_ms": 199.461
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 1,
          "response": "false",
          "task_id": "privacy",
          "tokens_per_second": 106.525,
          "wall_latency_ms": 196.685
        },
        {
          "output_tokens": 4,
          "passed": true,
          "repeat": 2,
          "response": "518",
          "task_id": "arithmetic",
          "tokens_per_second": 74.189,
          "wall_latency_ms": 161.358
        },
        {
          "output_tokens": 7,
          "passed": true,
          "repeat": 2,
          "response": "{\"kind\":\"missing_owner\"}",
          "task_id": "governance",
          "tokens_per_second": 63.745,
          "wall_latency_ms": 214.256
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 2,
          "response": "reject",
          "task_id": "health_policy",
          "tokens_per_second": 110.773,
          "wall_latency_ms": 123.018
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 2,
          "response": "B",
          "task_id": "budget_route",
          "tokens_per_second": 109.063,
          "wall_latency_ms": 123.17
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 2,
          "response": "false",
          "task_id": "privacy",
          "tokens_per_second": 110.632,
          "wall_latency_ms": 123.775
        }
      ],
      "total_samples": 10,
      "warm_p50_latency_ms": 196.685,
      "warm_p95_latency_ms": 286.334
    },
    {
      "cold_start_ms": 684.464,
      "context_tokens": 32768,
      "load_duration_ms": 635.984,
      "median_tokens_per_second": 303.26,
      "model": "qwen2.5:1.5b-instruct-q4_K_M",
      "model_size_bytes": 986061892,
      "parameter_size": "1.5B",
      "passed_samples": 8,
      "quality_accuracy": 0.8,
      "quantization_level": "Q4_K_M",
      "resident_memory_bytes": 1163080498,
      "samples": [
        {
          "output_tokens": 4,
          "passed": true,
          "repeat": 1,
          "response": "518",
          "task_id": "arithmetic",
          "tokens_per_second": 233.223,
          "wall_latency_ms": 158.956
        },
        {
          "output_tokens": 7,
          "passed": true,
          "repeat": 1,
          "response": "{\"kind\":\"missing_owner\"}",
          "task_id": "governance",
          "tokens_per_second": 206.125,
          "wall_latency_ms": 148.333
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 1,
          "response": "Reject",
          "task_id": "health_policy",
          "tokens_per_second": 332.889,
          "wall_latency_ms": 123.037
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 1,
          "response": "B",
          "task_id": "budget_route",
          "tokens_per_second": 328.03,
          "wall_latency_ms": 122.253
        },
        {
          "output_tokens": 2,
          "passed": false,
          "repeat": 1,
          "response": "true",
          "task_id": "privacy",
          "tokens_per_second": 320.359,
          "wall_latency_ms": 119.82
        },
        {
          "output_tokens": 4,
          "passed": true,
          "repeat": 2,
          "response": "518",
          "task_id": "arithmetic",
          "tokens_per_second": 227.247,
          "wall_latency_ms": 108.032
        },
        {
          "output_tokens": 7,
          "passed": true,
          "repeat": 2,
          "response": "{\"kind\":\"missing_owner\"}",
          "task_id": "governance",
          "tokens_per_second": 191.781,
          "wall_latency_ms": 128.496
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 2,
          "response": "Reject",
          "task_id": "health_policy",
          "tokens_per_second": 330.087,
          "wall_latency_ms": 99.901
        },
        {
          "output_tokens": 2,
          "passed": true,
          "repeat": 2,
          "response": "B",
          "task_id": "budget_route",
          "tokens_per_second": 318.016,
          "wall_latency_ms": 99.7
        },
        {
          "output_tokens": 2,
          "passed": false,
          "repeat": 2,
          "response": "true",
          "task_id": "privacy",
          "tokens_per_second": 303.26,
          "wall_latency_ms": 99.951
        }
      ],
      "total_samples": 10,
      "warm_p50_latency_ms": 119.82,
      "warm_p95_latency_ms": 158.956
    }
  ],
  "router_profiles": [
    {
      "context_tokens": 32768,
      "healthy": true,
      "input_cost_per_million": 0.0,
      "model_id": "qwen2.5:7b-instruct",
      "output_cost_per_million": 0.0,
      "p95_latency_ms": 286,
      "quality_score": 1.0,
      "resident_memory_mb": 4521
    },
    {
      "context_tokens": 32768,
      "healthy": true,
      "input_cost_per_million": 0.0,
      "model_id": "qwen2.5:1.5b-instruct-q4_K_M",
      "output_cost_per_million": 0.0,
      "p95_latency_ms": 159,
      "quality_score": 0.8,
      "resident_memory_mb": 1109
    }
  ],
  "suite": "modelopspilot-arm64-ollama-v1"
}
