evals/mbert/eval_results_arabic.json

{
    "eval_accuracy": 0.26391261358166496,
    "eval_loss": 3.8152461392225794,
    "eval_perplexity": 45.38792675716374
}