evals/alberti/eval_results_russian.json

{
    "eval_accuracy": 0.5853658536585366,
    "eval_loss": 2.086235640967887,
    "eval_perplexity": 8.054537852308378
}