{
  "generatedBy": "bench.mjs",
  "startedAt": "2026-10-05T06:22:44.098Z",
  "finishedAt": "2026-10-05T06:22:44.238Z",
  "machine": {
    "cpu": "Apple M1",
    "os": "Darwin 26.3",
    "arch": "arm64",
    "node": "v22.18.0",
    "python": "Python 3.13.5"
  },
  "metricCount": 28,
  "metrics": [
    {
      "metric": "llamacpp_build",
      "value": "b11351",
      "unit": "",
      "note": "llama-b11351-bin-macos-arm64.tar.gz, Metal"
    },
    {
      "metric": "model",
      "value": "Qwen/Qwen2.5-1.5B-Instruct-GGUF (official)",
      "unit": ""
    },
    {
      "metric": "dataset",
      "value": "wikitext-2-raw-v1 wiki.test.raw, sha256 173c87a5...",
      "unit": ""
    },
    {
      "metric": "kld_reference",
      "value": "Q8_0 (fp16 download incomplete)",
      "unit": ""
    },
    {
      "metric": "q8_0_file_bytes",
      "value": 1894532128,
      "unit": "bytes",
      "note": "downloaded file size"
    },
    {
      "metric": "q8_0_ppl_100chunks",
      "value": 10.3258,
      "unit": "",
      "note": "llama-perplexity -c 512 --chunks 100; +/- 0.17161",
      "n": 100
    },
    {
      "metric": "q8_0_pp512_tps",
      "value": 586.87,
      "unit": "tok/s",
      "note": "llama-bench mean of 3; samples [595.56, 584.879, 580.18]",
      "n": 3
    },
    {
      "metric": "q8_0_tg128_tps",
      "value": 31.71,
      "unit": "tok/s",
      "note": "llama-bench mean of 3; samples [32.3169, 32.1695, 30.6452]",
      "n": 3
    },
    {
      "metric": "q5_k_m_file_bytes",
      "value": 1285494304,
      "unit": "bytes",
      "note": "downloaded file size"
    },
    {
      "metric": "q5_k_m_ppl_100chunks",
      "value": 10.3987,
      "unit": "",
      "note": "llama-perplexity -c 512 --chunks 100; +/- 0.17275",
      "n": 100
    },
    {
      "metric": "q5_k_m_pp512_tps",
      "value": 500.14,
      "unit": "tok/s",
      "note": "llama-bench mean of 3; samples [504.516, 497.427, 498.473]",
      "n": 3
    },
    {
      "metric": "q5_k_m_tg128_tps",
      "value": 38.31,
      "unit": "tok/s",
      "note": "llama-bench mean of 3; samples [38.8876, 38.0663, 37.9707]",
      "n": 3
    },
    {
      "metric": "q4_k_m_file_bytes",
      "value": 1117320736,
      "unit": "bytes",
      "note": "downloaded file size"
    },
    {
      "metric": "q4_k_m_ppl_100chunks",
      "value": 10.7509,
      "unit": "",
      "note": "llama-perplexity -c 512 --chunks 100; +/- 0.17961",
      "n": 100
    },
    {
      "metric": "q4_k_m_pp512_tps",
      "value": 559.44,
      "unit": "tok/s",
      "note": "llama-bench mean of 3; samples [560.586, 558.887, 558.835]",
      "n": 3
    },
    {
      "metric": "q4_k_m_tg128_tps",
      "value": 50.12,
      "unit": "tok/s",
      "note": "llama-bench mean of 3; samples [49.6314, 50.1359, 50.5937]",
      "n": 3
    },
    {
      "metric": "q5_k_m_mean_kld_vs_q8_0",
      "value": 0.010713,
      "unit": "",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q5_k_m_median_kld_vs_q8_0",
      "value": 0.007513,
      "unit": "",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q5_k_m_p99_kld_vs_q8_0",
      "value": 0.067402,
      "unit": "",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q5_k_m_same_top_token_vs_q8_0",
      "value": 94.078,
      "unit": "%",
      "note": "share of positions where the top token matches Q8_0",
      "n": 20
    },
    {
      "metric": "q5_k_m_ppl_ratio_vs_q8_0",
      "value": 1.007042,
      "unit": "ratio",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q5_k_m_rms_delta_p",
      "value": 2.648,
      "unit": "%",
      "note": "RMS change in token probability vs Q8_0",
      "n": 20
    },
    {
      "metric": "q4_k_m_mean_kld_vs_q8_0",
      "value": 0.036127,
      "unit": "",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q4_k_m_median_kld_vs_q8_0",
      "value": 0.022159,
      "unit": "",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q4_k_m_p99_kld_vs_q8_0",
      "value": 0.254363,
      "unit": "",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q4_k_m_same_top_token_vs_q8_0",
      "value": 90.039,
      "unit": "%",
      "note": "share of positions where the top token matches Q8_0",
      "n": 20
    },
    {
      "metric": "q4_k_m_ppl_ratio_vs_q8_0",
      "value": 1.049701,
      "unit": "ratio",
      "note": "20 chunks",
      "n": 20
    },
    {
      "metric": "q4_k_m_rms_delta_p",
      "value": 5.12,
      "unit": "%",
      "note": "RMS change in token probability vs Q8_0",
      "n": 20
    }
  ],
  "log": [
    ""
  ]
}