eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -2.0315260887145996,
    "eval_logits/rejected": -2.0060105323791504,
    "eval_logps/chosen": -155.9133758544922,
    "eval_logps/rejected": -168.3367919921875,
    "eval_loss": 0.22132937610149384,
    "eval_rewards/accuracies": 0.7689999938011169,
    "eval_rewards/chosen": -3.684920072555542,
    "eval_rewards/margins": 2.584770917892456,
    "eval_rewards/rejected": -6.2696919441223145,
    "eval_runtime": 65.5407,
    "eval_samples_per_second": 15.258,
    "eval_steps_per_second": 1.907
}