eval_results.json

{
    "epoch": 3.0,
    "eval_log_odds_chosen": 0.26997339725494385,
    "eval_log_odds_ratio": -0.7197813391685486,
    "eval_logits/chosen": 284.18389892578125,
    "eval_logits/rejected": 254.83587646484375,
    "eval_logps/chosen": -1.2543175220489502,
    "eval_logps/rejected": -1.4255971908569336,
    "eval_loss": 1.7139935493469238,
    "eval_nll_loss": 1.6444751024246216,
    "eval_rewards/accuracies": 0.5467625856399536,
    "eval_rewards/chosen": -0.06271587312221527,
    "eval_rewards/margins": 0.008563979528844357,
    "eval_rewards/rejected": -0.0712798535823822,
    "eval_runtime": 25.9538,
    "eval_samples": 553,
    "eval_samples_per_second": 21.307,
    "eval_steps_per_second": 5.356
}