evals/eval_results_mbert.json

{
    "eval_accuracy": 0.2531256913599916,
    "eval_loss": 4.986973319130462,
    "eval_perplexity": 146.49236618254264
}