eval_results.json

{
    "epoch": 4.0,
    "eval_logits/chosen": -2.1607675552368164,
    "eval_logits/rejected": -2.156055450439453,
    "eval_logps/chosen": -33.63822555541992,
    "eval_logps/rejected": -37.278076171875,
    "eval_loss": 0.4762561619281769,
    "eval_rewards/accuracies": 0.5685215592384338,
    "eval_rewards/chosen": 0.27742961049079895,
    "eval_rewards/margins": 0.11044850200414658,
    "eval_rewards/rejected": 0.16698110103607178,
    "eval_runtime": 145.2249,
    "eval_samples": 343,
    "eval_samples_per_second": 2.362,
    "eval_steps_per_second": 0.296
}