evals/mbert/eval_results_french.json

{
    "eval_accuracy": 0.36305560250126756,
    "eval_loss": 3.776505989099206,
    "eval_perplexity": 43.66321515247085
}