eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -3.0115926265716553,
    "eval_logits/rejected": -3.022681474685669,
    "eval_logps/chosen": -266.7742004394531,
    "eval_logps/rejected": -253.80645751953125,
    "eval_loss": 0.6836252808570862,
    "eval_rewards/accuracies": 0.5225093364715576,
    "eval_rewards/chosen": -0.020729802548885345,
    "eval_rewards/margins": 0.020151691511273384,
    "eval_rewards/rejected": -0.040862053632736206,
    "eval_runtime": 165.8484,
    "eval_samples": 2000,
    "eval_samples_per_second": 12.21,
    "eval_steps_per_second": 0.386
}