evals/alberti/eval_results_english.json

{
    "eval_accuracy": 0.5320815459645918,
    "eval_loss": 2.394027611476076,
    "eval_perplexity": 10.957537892911233
}