eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.15903639793396,
    "eval_logits/rejected": -2.1050093173980713,
    "eval_logps/chosen": -263.8238525390625,
    "eval_logps/rejected": -217.84848022460938,
    "eval_loss": 0.498769074678421,
    "eval_rewards/accuracies": 0.8500000238418579,
    "eval_rewards/chosen": 0.9567789435386658,
    "eval_rewards/margins": 2.93479585647583,
    "eval_rewards/rejected": -1.9780166149139404,
    "eval_runtime": 124.8107,
    "eval_samples_per_second": 15.223,
    "eval_steps_per_second": 0.24
}