{
  "recordKind": "local measured comparison; not a general quality ranking",
  "runtime": {
    "vendor": "official ggml-org/llama.cpp release",
    "tag": "b11576",
    "commit": "358a420ec4ed59b6b0be08b1d1f71dd478b99a99",
    "backend": "CUDA12.4",
    "gpu": "RTX5090",
    "cpu": "Ryzen9 9950X",
    "systemRamBytes": 66125668352,
    "driver": "617.42"
  },
  "models": [
    {
      "repo": "XHToken/Spark-X2.5-1.7B-GGUF",
      "revision": "1f7fa33b1245c14730da39e125714ad3a327901b",
      "quantization": "Q4_K_M",
      "file": "Spark-X2.5-1.7B-Q4_K_M.gguf",
      "bytes": 1107457856,
      "sha256": "902bde2522394954ac17821b3e5fd0df02defbc6944f122253f2580acf0503f4"
    },
    {
      "repo": "XHToken/Spark-X2.5-4B-GGUF",
      "revision": "9826e0be84e6e6e8b9668abc91421109a1df1e2d",
      "quantization": "Q4_K_M",
      "file": "Spark-X2.5-4B-Q4_K_M.gguf",
      "bytes": 2600224352,
      "sha256": "adfcfa19a4ed6a5985da8bf565fe15f8e1a7e131d79bae2d19d48d1c40109428"
    }
  ],
  "conditions": {
    "context": 1024,
    "initialMaxOutput": 64,
    "followupMaxOutput": 512,
    "seed": 20261025,
    "temperature": 0,
    "defaultReasoning": true,
    "timeoutSecondsPerProcess": 120
  },
  "memoryScope": "sampled process working set and whole GPU including desktop; not true peaks/model requirements",
  "timeScope": "whole process elapsed includes startup/loading/cache/JIT/reasoning; not TTFT or a cold/warm speed comparison",
  "rubric": {
    "reviewer": "active Codex agent applying explicit rubric; human review pending",
    "scope": "three fixed synthetic Japanese questions, not1M context or general quality benchmark",
    "questions": [
      {
        "id": "arithmetic",
        "prompt": "17と25を足した結果を、数字だけで答えてください。",
        "rubric": "Answer must be42 (allow whitespace/punctuation only); any other number or incorrect arithmetic fails.",
        "expected": "42"
      },
      {
        "id": "extraction",
        "prompt": "架空の作業メモ: 田中さんは10月30日までに画像を3枚用意します。担当者と期限だけを日本語で答えてください。",
        "rubric": "Must identify田中さん and10月30日; no conflicting person/date. Extra task text recorded as instruction deviation separately.",
        "expected": [
          "田中",
          "10月30日"
        ]
      },
      {
        "id": "sorting",
        "prompt": "数字の9、2、5を小さい順に並べ、数字と読点だけで答えてください。",
        "rubric": "Numeric sequence must be2,5,9; allow Japanese/ASCII separators and whitespace. Extra numbers or reordered sequence fail.",
        "expected": [
          2,
          5,
          9
        ]
      }
    ]
  },
  "rows": [
    {
      "model": "spark1.7b",
      "question": "arithmetic",
      "finalAnswer": "42",
      "finalAnswerReached": true,
      "outputTokenLimit": 512,
      "generatedTokens": 117,
      "stopReason": "eos_or_stop",
      "instructionDeviation": false,
      "elapsedSecondsIncludingProcessStart": 1.610208000000057,
      "sampledMaxWorkingSetBytes": 1608458240,
      "sampledMaxWholeGpuMiB": 3512,
      "offloadedLayers": "29/29",
      "rubricPassed": true,
      "arithmeticResultCorrect": true,
      "measuredAt": "2026-10-25T17:29:43.193780+09:00",
      "input": "17と25を足した結果を、数字だけで答えてください。",
      "provenance": [
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-arithmetic/result.json",
          "sha256": "15c41c459030eaeee90bc06a95cad3e3bf0b963333ba427d643bf464d252970b"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-arithmetic/stdout.log",
          "sha256": "29550fcd37202113f0dbf504553edcb796ebf3c927ce527b108a7874f176d248"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-arithmetic/engine.log",
          "sha256": "f5a616ccb364b84e686eacaf9924efeb7f3a6a063ad2caa93c4b4fda358a21ae"
        }
      ],
      "original64TokenTrial": {
        "status": "output_limit_before_final_answer",
        "elapsedSecondsIncludingProcessStart": 23.8141910000013,
        "provenance": [
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-arithmetic/result.json",
            "sha256": "940e681c4b36138bc5a5deaab06d87e2b840c74e509b4a78112f124e17565c0d"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-arithmetic/stdout.log",
            "sha256": "131374ee8d9c9f994291ec1da4a64b95b1e21f42dce78645341d6c1eb1053622"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-arithmetic/engine.log",
            "sha256": "f5dad9299f4b90f637fabee4df740e261cc7e0d6a4977e2bebfaa407f87ceb3c"
          }
        ]
      }
    },
    {
      "model": "spark1.7b",
      "question": "extraction",
      "finalAnswer": "**作業メモの回答**\n\n**期限：** 10月30日まで\n\n**数量：** 画像を3枚用意する\n\n（注：本回答は期限と数量のみを日本語で伝えるものです。）",
      "finalAnswerReached": true,
      "outputTokenLimit": 512,
      "generatedTokens": 391,
      "stopReason": "eos_or_stop",
      "instructionDeviation": true,
      "elapsedSecondsIncludingProcessStart": 2.3969746000002488,
      "sampledMaxWorkingSetBytes": 1610100736,
      "sampledMaxWholeGpuMiB": 3516,
      "offloadedLayers": "29/29",
      "rubricPassed": false,
      "measuredAt": "2026-10-25T17:29:47.364513+09:00",
      "input": "架空の作業メモ: 田中さんは10月30日までに画像を3枚用意します。担当者と期限だけを日本語で答えてください。",
      "provenance": [
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-extraction/result.json",
          "sha256": "3dd19628229c296ed7d2fe4fd208f686e3d862b492c2f8817241970010427323"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-extraction/stdout.log",
          "sha256": "314ffc5d35915c49682ce161902dd47cad03d9eff292e7c85fbf500ea368fcf9"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-extraction/engine.log",
          "sha256": "24ebc2b9edeaa5fc5b031f72f2af45c03ea1ca5f5ea5144ef8060ac41870f450"
        }
      ],
      "original64TokenTrial": {
        "status": "output_limit_before_final_answer",
        "elapsedSecondsIncludingProcessStart": 1.5579614999987825,
        "provenance": [
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-extraction/result.json",
            "sha256": "20eb06c7c484227985aa9436d792440ecb34a07ea8f02943c8fea046a181d0f2"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-extraction/stdout.log",
            "sha256": "6cc9dd428400f5fe8dcdb653c612252d8b87c7e5dd0b5b8dca224600dcccb363"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-extraction/engine.log",
            "sha256": "670b65d649ad31474522b8553f9c356b4d74ed85a770cf97489cd1a711635078"
          }
        ]
      }
    },
    {
      "model": "spark1.7b",
      "question": "sorting",
      "finalAnswer": "2、5、9",
      "finalAnswerReached": true,
      "outputTokenLimit": 512,
      "generatedTokens": 251,
      "stopReason": "eos_or_stop",
      "instructionDeviation": false,
      "elapsedSecondsIncludingProcessStart": 1.995171799999298,
      "sampledMaxWorkingSetBytes": 1608617984,
      "sampledMaxWholeGpuMiB": 3623,
      "offloadedLayers": "29/29",
      "rubricPassed": true,
      "measuredAt": "2026-10-25T17:29:52.341556+09:00",
      "input": "数字の9、2、5を小さい順に並べ、数字と読点だけで答えてください。",
      "provenance": [
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-sorting/result.json",
          "sha256": "fbfc84249864d915eb1338c6316247b46905075d05934578b1a031ed47af2a27"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-sorting/stdout.log",
          "sha256": "e870c8513c69b2a88fe51b63400f4a9d0746c84bfa828a7786386b599ac0495c"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark1.7b-sorting/engine.log",
          "sha256": "47e7254d5ce83babdf1adc7c591a065f2cafc7ac7578a7b00cb5b6d66194255b"
        }
      ],
      "original64TokenTrial": {
        "status": "output_limit_before_final_answer",
        "elapsedSecondsIncludingProcessStart": 1.5320941999998468,
        "provenance": [
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-sorting/result.json",
            "sha256": "b3b45cf50b82018638938f24e39a050ed29248cbcbf7178e530ebe1c2fd41df0"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-sorting/stdout.log",
            "sha256": "473b64e32884df2af3df50eb889a593ae968a7b39a5ab5c9371cdf6bc9e4905d"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark1.7b-sorting/engine.log",
            "sha256": "a32a79ae6c712f7a4d638d762440ea6cdba34a6aa23ee88d93e547ecb6a5728b"
          }
        ]
      }
    },
    {
      "model": "spark4b",
      "question": "arithmetic",
      "finalAnswer": "17 + 25 = 42\n\n答え：42",
      "finalAnswerReached": true,
      "outputTokenLimit": 512,
      "generatedTokens": 176,
      "stopReason": "eos_or_stop",
      "instructionDeviation": true,
      "elapsedSecondsIncludingProcessStart": 2.5206292999992,
      "sampledMaxWorkingSetBytes": 3112460288,
      "sampledMaxWholeGpuMiB": 5055,
      "offloadedLayers": "37/37",
      "rubricPassed": false,
      "arithmeticResultCorrect": true,
      "measuredAt": "2026-10-25T17:29:57.698847+09:00",
      "input": "17と25を足した結果を、数字だけで答えてください。",
      "provenance": [
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-arithmetic/result.json",
          "sha256": "7cbda1a62eb5dc95c4aeffff357fd70ab1f37f4f6f06e9c7017ba6989cec2e1f"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-arithmetic/stdout.log",
          "sha256": "96f53ed0cae1db48d31c203dfe0432f93d35a76e6bcf0b613abd935ec54a390e"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-arithmetic/engine.log",
          "sha256": "b3696b3598faf8cacd8b75f9e97af6f4be1c26a71731863cda06cad87068d892"
        }
      ],
      "original64TokenTrial": {
        "status": "output_limit_before_final_answer",
        "elapsedSecondsIncludingProcessStart": 2.1303333999985625,
        "provenance": [
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-arithmetic/result.json",
            "sha256": "82a189f962677db5b3f66eae81a023ae6a9b85fe96f04d6263b3e37294e52895"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-arithmetic/stdout.log",
            "sha256": "d62f15d013005f04167c5ab1675756f7c0bd0936b467bd211c5f5f7fa03856f4"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-arithmetic/engine.log",
            "sha256": "d570b12e9aa70da3448cb6ead09466c5cc75d5448b304dd4bd4918a2cd22cc12"
          }
        ]
      }
    },
    {
      "model": "spark4b",
      "question": "extraction",
      "finalAnswer": "担当者：田中さん  \n期限：10月30日",
      "finalAnswerReached": true,
      "outputTokenLimit": 512,
      "generatedTokens": 394,
      "stopReason": "eos_or_stop",
      "instructionDeviation": false,
      "elapsedSecondsIncludingProcessStart": 3.672395100000358,
      "sampledMaxWorkingSetBytes": 3116285952,
      "sampledMaxWholeGpuMiB": 5059,
      "offloadedLayers": "37/37",
      "rubricPassed": true,
      "measuredAt": "2026-10-25T17:30:04.590936+09:00",
      "input": "架空の作業メモ: 田中さんは10月30日までに画像を3枚用意します。担当者と期限だけを日本語で答えてください。",
      "provenance": [
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-extraction/result.json",
          "sha256": "bb51c5bbbaa9ca9963af996ebdacb49691cbb234aae8e8ec2af1f132fa689f4d"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-extraction/stdout.log",
          "sha256": "46586bde8a660686484de58397d5d3531a9ff53627b65e6bedadb07e3ac0cebd"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-extraction/engine.log",
          "sha256": "83d4a9bbe390f06eb78b703ee5fe6e52ce466d3a9aa2beb26110eb68c3b99f19"
        }
      ],
      "original64TokenTrial": {
        "status": "output_limit_before_final_answer",
        "elapsedSecondsIncludingProcessStart": 2.346837200000664,
        "provenance": [
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-extraction/result.json",
            "sha256": "8ee257f20b3b5d1afd9c0422e842cc16c2d62f5d8364d4ef45a499e6cd589283"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-extraction/stdout.log",
            "sha256": "d522c39c5a6304552a3fd9b177e22097d1066a4e9da86211e0efe9fa4af17087"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-extraction/engine.log",
            "sha256": "329a195be0bc0bdd21acd4fa83e5fb9837e721222e5d11564a78667fbb5b051f"
          }
        ]
      }
    },
    {
      "model": "spark4b",
      "question": "sorting",
      "finalAnswer": "2、5、9",
      "finalAnswerReached": true,
      "outputTokenLimit": 512,
      "generatedTokens": 222,
      "stopReason": "eos_or_stop",
      "instructionDeviation": false,
      "elapsedSecondsIncludingProcessStart": 2.9524225000004662,
      "sampledMaxWorkingSetBytes": 3113680896,
      "sampledMaxWholeGpuMiB": 5055,
      "offloadedLayers": "37/37",
      "rubricPassed": true,
      "measuredAt": "2026-10-25T17:30:12.578105+09:00",
      "input": "数字の9、2、5を小さい順に並べ、数字と読点だけで答えてください。",
      "provenance": [
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-sorting/result.json",
          "sha256": "e809cae52d811a3ad16bd8e3874a55c8a5ea484296d85f4a60c57c2be37f7d7a"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-sorting/stdout.log",
          "sha256": "84073c5d70e0a537caee732e6ccb6293ad66afa6ff35f3d83c8cb777beb3d305"
        },
        {
          "path": "research/all100-20261025/d-spark-pair-512/spark4b-sorting/engine.log",
          "sha256": "1627a16c07834640e7516ea90fea375793895915eae40049b7e8840cdf858f5d"
        }
      ],
      "original64TokenTrial": {
        "status": "output_limit_before_final_answer",
        "elapsedSecondsIncludingProcessStart": 2.3548025000000052,
        "provenance": [
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-sorting/result.json",
            "sha256": "dcdf32ac475d0c2884d8c604f7c181ea4413bc4d18209e31a02550eff4fbec98"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-sorting/stdout.log",
            "sha256": "5482dcfa4d45c3f9d9ead9126047d6b77c3c6d59a7e2ca3654d582be16fe54b6"
          },
          {
            "path": "research/all100-20261025/d-spark-pair/spark4b-sorting/engine.log",
            "sha256": "07433547d681900440ee96b8125c75cdbe9b449ef1ff8065f30dd293081454fe"
          }
        ]
      }
    }
  ]
}