eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -3.799543857574463,
    "eval_logits/rejected": -3.8772289752960205,
    "eval_logps/chosen": -599.8087768554688,
    "eval_logps/rejected": -667.6969604492188,
    "eval_loss": 0.42904943227767944,
    "eval_rewards/accuracies": 0.7680000066757202,
    "eval_rewards/chosen": -2.1400206089019775,
    "eval_rewards/margins": 1.2330021858215332,
    "eval_rewards/rejected": -3.37302303314209,
    "eval_runtime": 203.6811,
    "eval_samples": 2000,
    "eval_samples_per_second": 9.819,
    "eval_steps_per_second": 0.614
}