all_results.json

{
    "epoch": 2.999098751126561,
    "eval_log_ratio_diff/mean": 5.0,
    "eval_logits/chosen": -1.65625,
    "eval_logits/rejected": -1.671875,
    "eval_logps/chosen": -324.0,
    "eval_logps/rejected": -304.0,
    "eval_loss": 0.6349183320999146,
    "eval_nll_loss": 1.078125,
    "eval_ref_probs/chosen": 0.0038271306548267603,
    "eval_ref_probs/rejected": 0.0034569373819977045,
    "eval_rejected_term/max": -0.0005586988409049809,
    "eval_rejected_term/min": -0.0005586988409049809,
    "eval_rewards/accuracies": 0.7160000205039978,
    "eval_rewards/chosen": -0.25,
    "eval_rewards/margins": 0.5,
    "eval_rewards/rejected": -0.75,
    "eval_runtime": 25.3311,
    "eval_samples_per_second": 39.477,
    "eval_steps_per_second": 9.869
}