aggregate_metrics.json

{
  "test_macro_f1_mean": 0.8679318276840338,
  "test_macro_f1_std": 0.00586273502560902,
  "test_macro_f1_text": "0.8679 ± 0.0059",
  "test_accuracy_mean": 0.8678637200736649,
  "test_accuracy_std": 0.00558211586233062,
  "test_accuracy_text": "0.8679 ± 0.0056",
  "test_macro_precision_mean": 0.8683751525788259,
  "test_macro_precision_std": 0.005480868817224346,
  "test_macro_precision_text": "0.8684 ± 0.0055",
  "test_macro_recall_mean": 0.8678752996645088,
  "test_macro_recall_std": 0.006121784554093659,
  "test_macro_recall_text": "0.8679 ± 0.0061",
  "dev_macro_f1_mean": 0.874644258611763,
  "dev_macro_f1_std": 0.0013938767762517485,
  "dev_macro_f1_text": "0.8746 ± 0.0014",
  "task": "ViFactCheck-gold-evidence",
  "dataset": "ViFactCheck",
  "model_key": "velectra",
  "model_name": "vELECTRA-base",
  "base_model": "FPTAI/velectra-base-discriminator-cased",
  "seeds": [
    22,
    42,
    202
  ],
  "representative_seed": 42,
  "selection_rule": "maximum development Macro-F1; seed ascending tie-break",
  "split_policy": "merged_stratified_80_10_10",
  "split_seed": 42,
  "max_length": 256,
  "epochs": 3,
  "effective_batch_size": 8
}