eval_results.json

{
    "eval_accuracy": 0.6890367604912527,
    "eval_loss": 1.4896034919686423,
    "eval_perplexity": 4.435336523745398
}