eval_results.json

{
    "epoch": 2.999098751126561,
    "eval_kl/chosen": 0.00024712123558856547,
    "eval_kl/rejected": 0.0003913869149982929,
    "eval_logits/chosen": -1.421875,
    "eval_logits/rejected": -1.390625,
    "eval_logps/chosen": -322.0,
    "eval_logps/rejected": -298.0,
    "eval_loss": 0.8087028861045837,
    "eval_nll_loss": 1.046875,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": 0.0003913869149982929,
    "eval_rejected_term/min": 0.0003913869149982929,
    "eval_rewards/accuracies": 0.6200000047683716,
    "eval_rewards/chosen": -0.0034942626953125,
    "eval_rewards/margins": 0.1416015625,
    "eval_rewards/rejected": -0.1455078125,
    "eval_runtime": 25.2817,
    "eval_samples_per_second": 39.554,
    "eval_steps_per_second": 9.889
}