eval_results.json

{
    "eval_accuracy": 0.6885889057979672,
    "eval_loss": 1.430497475427537,
    "eval_perplexity": 4.180778509252052
}