training_summary.json

{
  "best_epoch": 26,
  "best_macro_f1": 0.9085411885232777,
  "epochs_run": 30,
  "history_path": "training_history.json"
}