evals/alberti/eval_results_french.json

{
    "eval_accuracy": 0.5842149738042928,
    "eval_loss": 2.0680841484282575,
    "eval_perplexity": 7.909654867720871
}