aggregate_metrics.json

{
  "test_macro_f1_mean": 0.7859431230658753,
  "test_macro_f1_std": 0.0078090422016252135,
  "test_macro_f1_text": "0.7859 ± 0.0078",
  "test_accuracy_mean": 0.7863720073664825,
  "test_accuracy_std": 0.008439365920728966,
  "test_accuracy_text": "0.7864 ± 0.0084",
  "test_macro_precision_mean": 0.7861867400362637,
  "test_macro_precision_std": 0.007303026040786383,
  "test_macro_precision_text": "0.7862 ± 0.0073",
  "test_macro_recall_mean": 0.7869905109603467,
  "test_macro_recall_std": 0.008424459089079234,
  "test_macro_recall_text": "0.7870 ± 0.0084",
  "dev_macro_f1_mean": 0.7795388436295254,
  "dev_macro_f1_std": 0.012688247976241145,
  "dev_macro_f1_text": "0.7795 ± 0.0127",
  "task": "ViFactCheck-gold-evidence",
  "dataset": "ViFactCheck",
  "model_key": "wikibert",
  "model_name": "WikiBERT",
  "base_model": "TurkuNLP/wikibert-base-vi-cased",
  "seeds": [
    22,
    42,
    202
  ],
  "representative_seed": 202,
  "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
}