eval_results.json

{
    "epoch": 2.999098751126561,
    "eval_kl/chosen": 0.0008587873307988048,
    "eval_kl/rejected": -0.00045978292473591864,
    "eval_logits/chosen": -1.125,
    "eval_logits/rejected": -1.1015625,
    "eval_logps/chosen": -330.0,
    "eval_logps/rejected": -302.0,
    "eval_loss": 1.2715272903442383,
    "eval_nll_loss": 1.078125,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": -0.00045978292473591864,
    "eval_rejected_term/min": -0.00045978292473591864,
    "eval_rewards/accuracies": 0.30000001192092896,
    "eval_rewards/chosen": -0.88671875,
    "eval_rewards/margins": -0.453125,
    "eval_rewards/rejected": -0.435546875,
    "eval_runtime": 25.5569,
    "eval_samples_per_second": 39.128,
    "eval_steps_per_second": 9.782
}