eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -3.394435405731201,
    "eval_logits/rejected": -3.4397640228271484,
    "eval_logps/chosen": -560.387451171875,
    "eval_logps/rejected": -564.275390625,
    "eval_loss": 0.7548772692680359,
    "eval_rewards/accuracies": 0.46875,
    "eval_rewards/chosen": 3.891627311706543,
    "eval_rewards/diff": -0.19507551193237305,
    "eval_rewards/diff_abs": 1.0197707414627075,
    "eval_rewards/rejected": 3.839632511138916,
    "eval_rewards/student_margin": 0.05199478566646576,
    "eval_rewards/teacher_margin": 0.2470703125,
    "eval_runtime": 26.2432,
    "eval_samples": 1543,
    "eval_samples_per_second": 58.796,
    "eval_steps_per_second": 0.152
}