eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.7826476097106934,
    "eval_logits/rejected": -2.672537326812744,
    "eval_logps/chosen": -293.99981689453125,
    "eval_logps/rejected": -279.5479736328125,
    "eval_loss": 0.6910352110862732,
    "eval_rewards/accuracies": 0.8125,
    "eval_rewards/chosen": -3.9218149185180664,
    "eval_rewards/margins": 4.372367858886719,
    "eval_rewards/rejected": -8.294181823730469,
    "eval_runtime": 43.6649,
    "eval_samples": 2000,
    "eval_samples_per_second": 45.803,
    "eval_steps_per_second": 0.366
}