{
  "study": "Llama3.1-8B to Qwen2.5-72B-Instruct mapped KV, October 10 validation",
  "provenance": {
    "experiment_artifact": "results/onetriangle-model-images-20261010/qwen72b-final-v1/benchmark/summary.json",
    "sha256": "145978ab430a916ebbed104b3e1ab65dcc192c0d6315a31c83a476c0df926d6c"
  },
  "measurement": "complete sequential HTTP request latency; not streamed TTFT or throughput",
  "hardware": "3 H100 80 GB: 1 source, 2 BF16 target; warm resident models",
  "sample": {
    "independent_fact_sets": 4,
    "distinct_fact_set_length_cases": 12,
    "repetitions_per_mode": 3,
    "measured_requests": 72,
    "warmups_excluded": 2
  },
  "limits": [
    "Synthetic box-color recall only.",
    "Includes source work and generation; excludes model loading.",
    "No production throughput, billing, or general quality claim."
  ],
  "groups": [
    {
      "accuracy": 1.0,
      "correct": 12,
      "http_200": 12,
      "median_http_ms": 126.16084422916174,
      "mode": "normal",
      "nominal_context_tokens": 512,
      "requests": 12
    },
    {
      "accuracy": 1.0,
      "correct": 12,
      "http_200": 12,
      "median_http_ms": 172.00011294335127,
      "mode": "onetriangle",
      "nominal_context_tokens": 512,
      "requests": 12
    },
    {
      "accuracy": 1.0,
      "correct": 12,
      "http_200": 12,
      "median_http_ms": 332.91084319353104,
      "mode": "normal",
      "nominal_context_tokens": 2048,
      "requests": 12
    },
    {
      "accuracy": 1.0,
      "correct": 12,
      "http_200": 12,
      "median_http_ms": 242.1599766239524,
      "mode": "onetriangle",
      "nominal_context_tokens": 2048,
      "requests": 12
    },
    {
      "accuracy": 1.0,
      "correct": 12,
      "http_200": 12,
      "median_http_ms": 1058.080531656742,
      "mode": "normal",
      "nominal_context_tokens": 7500,
      "requests": 12
    },
    {
      "accuracy": 1.0,
      "correct": 12,
      "http_200": 12,
      "median_http_ms": 581.1426187865436,
      "mode": "onetriangle",
      "nominal_context_tokens": 7500,
      "requests": 12
    }
  ],
  "latency_ratios": [
    {
      "nominal_context_tokens": 512,
      "normal_over_onetriangle_http_ratio": 0.7334927987559704
    },
    {
      "nominal_context_tokens": 2048,
      "normal_over_onetriangle_http_ratio": 1.3747558446064136
    },
    {
      "nominal_context_tokens": 7500,
      "normal_over_onetriangle_http_ratio": 1.820689960523064
    }
  ],
  "all_exact_normalized_correct": true
}
