eval_results.json

{
    "epoch": 3.0,
    "eval_log_odds_chosen": 0.3720725178718567,
    "eval_log_odds_ratio": -0.699243426322937,
    "eval_logits/chosen": 295.62017822265625,
    "eval_logits/rejected": 266.0260314941406,
    "eval_logps/chosen": -1.3004335165023804,
    "eval_logps/rejected": -1.5281699895858765,
    "eval_loss": 1.7558966875076294,
    "eval_nll_loss": 1.6941194534301758,
    "eval_rewards/accuracies": 0.597122311592102,
    "eval_rewards/chosen": -0.06502167135477066,
    "eval_rewards/margins": 0.011386833153665066,
    "eval_rewards/rejected": -0.0764085054397583,
    "eval_runtime": 90.3714,
    "eval_samples": 553,
    "eval_samples_per_second": 6.119,
    "eval_steps_per_second": 1.538
}