evaluation.json

{
  "summary": {
    "task_accuracy": 0.9121989121989122,
    "task_macro_f1": 0.9103734961055048,
    "complexity_accuracy": 0.8679098679098679,
    "complexity_macro_f1": 0.8723688079771166,
    "training_seconds": 13.171398878097534
  },
  "task_report": {
    "general_chat": {
      "precision": 0.9310344827586207,
      "recall": 0.9204545454545454,
      "f1-score": 0.9257142857142857,
      "support": 176.0
    },
    "writing": {
      "precision": 0.974025974025974,
      "recall": 0.9566326530612245,
      "f1-score": 0.9652509652509652,
      "support": 392.0
    },
    "translation": {
      "precision": 1.0,
      "recall": 1.0,
      "f1-score": 1.0,
      "support": 4.0
    },
    "summarization": {
      "precision": 0.9655172413793104,
      "recall": 0.9655172413793104,
      "f1-score": 0.9655172413793104,
      "support": 29.0
    },
    "research": {
      "precision": 0.8372093023255814,
      "recall": 0.8944099378881988,
      "f1-score": 0.8648648648648649,
      "support": 161.0
    },
    "coding": {
      "precision": 0.9132231404958677,
      "recall": 0.9057377049180327,
      "f1-score": 0.9094650205761317,
      "support": 244.0
    },
    "mathematics": {
      "precision": 0.8641975308641975,
      "recall": 0.8536585365853658,
      "f1-score": 0.8588957055214724,
      "support": 82.0
    },
    "document_analysis": {
      "precision": 0.8165137614678899,
      "recall": 0.9081632653061225,
      "f1-score": 0.8599033816425121,
      "support": 98.0
    },
    "high_stakes": {
      "precision": 0.8901098901098901,
      "recall": 0.801980198019802,
      "f1-score": 0.84375,
      "support": 101.0
    },
    "accuracy": 0.9121989121989122,
    "macro avg": {
      "precision": 0.9102034803808147,
      "recall": 0.9118393425125113,
      "f1-score": 0.9103734961055048,
      "support": 1287.0
    },
    "weighted avg": {
      "precision": 0.913815881736539,
      "recall": 0.9121989121989122,
      "f1-score": 0.912490699441833,
      "support": 1287.0
    }
  },
  "complexity_report": {
    "low": {
      "precision": 0.9166666666666666,
      "recall": 0.8856476079346558,
      "f1-score": 0.9008902077151335,
      "support": 857.0
    },
    "medium": {
      "precision": 0.7555555555555555,
      "recall": 0.8225806451612904,
      "f1-score": 0.7876447876447876,
      "support": 372.0
    },
    "high": {
      "precision": 0.9629629629629629,
      "recall": 0.896551724137931,
      "f1-score": 0.9285714285714286,
      "support": 58.0
    },
    "accuracy": 0.8679098679098679,
    "macro avg": {
      "precision": 0.878395061728395,
      "recall": 0.8682599924112924,
      "f1-score": 0.8723688079771166,
      "support": 1287.0
    },
    "weighted avg": {
      "precision": 0.8721848110736999,
      "recall": 0.8679098679098679,
      "f1-score": 0.8694047489299714,
      "support": 1287.0
    }
  },
  "task_confusion_matrix": [
    [
      162,
      2,
      0,
      0,
      1,
      3,
      4,
      0,
      4
    ],
    [
      1,
      375,
      0,
      0,
      5,
      10,
      1,
      0,
      0
    ],
    [
      0,
      0,
      4,
      0,
      0,
      0,
      0,
      0,
      0
    ],
    [
      0,
      0,
      0,
      28,
      0,
      0,
      1,
      0,
      0
    ],
    [
      4,
      1,
      0,
      1,
      144,
      3,
      0,
      5,
      3
    ],
    [
      3,
      5,
      0,
      0,
      6,
      221,
      0,
      6,
      3
    ],
    [
      0,
      0,
      0,
      0,
      4,
      2,
      70,
      6,
      0
    ],
    [
      4,
      0,
      0,
      0,
      5,
      0,
      0,
      89,
      0
    ],
    [
      0,
      2,
      0,
      0,
      7,
      3,
      5,
      3,
      81
    ]
  ],
  "complexity_confusion_matrix": [
    [
      759,
      98,
      0
    ],
    [
      64,
      306,
      2
    ],
    [
      5,
      1,
      52
    ]
  ]
}