training_summary.json

{
  "model_name": "google/byt5-small",
  "dataset_name": "wnkh/medieval-latin-ocr-correction",
  "train_samples": 90163,
  "val_samples": 4746,
  "training_time_hours": 3.8745588700638876,
  "final_train_loss": 0.4137230705283639,
  "final_eval_results": {
    "eval_loss": 0.36988574266433716,
    "eval_cer": 0.1665792765941789,
    "eval_cer_raw": 0.18652676107827693,
    "eval_exact_match": 0.1571849978929625,
    "eval_runtime": 1012.6473,
    "eval_samples_per_second": 4.687,
    "eval_steps_per_second": 1.172,
    "epoch": 7.0
  }
}