eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -16.75,
    "eval_logits/rejected": -16.5,
    "eval_logps/chosen": -880.0,
    "eval_logps/rejected": -900.0,
    "eval_loss": 1.452078104019165,
    "eval_rewards/accuracies": 0.521484375,
    "eval_rewards/chosen": -5.59375,
    "eval_rewards/margins": 0.5078125,
    "eval_rewards/rejected": -6.125,
    "eval_runtime": 46.5435,
    "eval_samples": 2000,
    "eval_samples_per_second": 42.971,
    "eval_steps_per_second": 0.688
}