all_results.json

{
    "epoch": 3.0,
    "eval_accuracy": 0.887,
    "eval_classification_report": "              precision    recall  f1-score   support\n0              0.924013  0.927063  0.925535  3030.000\n1              0.769792  0.761856  0.765803   970.000\naccuracy       0.887000  0.887000  0.887000     0.887\nmacro avg      0.846902  0.844459  0.845669  4000.000\nweighted avg   0.886614  0.887000  0.886800  4000.000",
    "eval_confusion_matrix": "[[2809  221]\n [ 231  739]]",
    "eval_confusion_matrix_norm": "[[0.92706271 0.07293729]\n [0.23814433 0.76185567]]",
    "eval_f1": 0.7658031088082901,
    "eval_f1_macro": 0.8456692644535684,
    "eval_f1_weighted": 0.8868003346108867,
    "eval_loss": 0.42191165685653687,
    "eval_runtime": 9.6157,
    "eval_samples": 4000,
    "eval_samples_per_second": 415.988,
    "eval_steps_per_second": 25.999,
    "train_loss": 0.25008199564615885,
    "train_runtime": 436.7193,
    "train_samples": 16000,
    "train_samples_per_second": 109.91,
    "train_steps_per_second": 6.869
}