training/eval_report.json

{
  "n_per_agent": 120,
  "base": "qwen2.5:7b-instruct",
  "ft": "finsight-qwen",
  "judge": "qwen3:8b",
  "scores": {
    "base": {
      "sentiment": {
        "n": 120,
        "accuracy": 0.275,
        "macro_f1": 0.3291,
        "parse_fail": 61
      },
      "event": {
        "n": 120,
        "rougeL": 0.0821
      },
      "summary": {
        "n": 120,
        "rougeL": 0.2636
      },
      "topic": {
        "n": 120,
        "rougeL": 0.1107,
        "finnl_acc": 0.7045,
        "finnl_n": 44
      },
      "qa": {
        "n": 120,
        "rougeL": 0.2181
      }
    },
    "ft": {
      "sentiment": {
        "n": 120,
        "accuracy": 0.7417,
        "macro_f1": 0.7286,
        "parse_fail": 0
      },
      "event": {
        "n": 120,
        "rougeL": 0.796
      },
      "summary": {
        "n": 120,
        "rougeL": 0.5444
      },
      "topic": {
        "n": 120,
        "rougeL": 0.5432,
        "finnl_acc": 0.8636,
        "finnl_n": 44
      },
      "qa": {
        "n": 120,
        "rougeL": 0.3176
      }
    }
  },
  "qa_judge": {
    "qa_n": 120,
    "comparisons": 240,
    "ft_win": 50,
    "base_win": 188,
    "tie": 2,
    "ft_win_rate": 0.2083
  }
}