evals/alberti/eval_results_german.json

{
    "eval_accuracy": 0.5835830474456813,
    "eval_loss": 1.9559251674484819,
    "eval_perplexity": 7.070457355700252
}