eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": 0.5663220286369324,
    "eval_logits/rejected": 1.4431953430175781,
    "eval_logps/chosen": -420.1173095703125,
    "eval_logps/rejected": -547.8262329101562,
    "eval_loss": 0.523222804069519,
    "eval_rewards/accuracies": 0.765625,
    "eval_rewards/chosen": -1.574873685836792,
    "eval_rewards/margins": 1.276767373085022,
    "eval_rewards/rejected": -2.8516414165496826,
    "eval_runtime": 104.8484,
    "eval_samples": 2000,
    "eval_samples_per_second": 19.075,
    "eval_steps_per_second": 0.305
}