eval_results.json

{
    "epoch": 32.0,
    "eval_logits/chosen": -0.3564789891242981,
    "eval_logits/rejected": -0.32587698101997375,
    "eval_logps/chosen": -1.786582350730896,
    "eval_logps/rejected": -1.8725693225860596,
    "eval_loss": 0.7796740531921387,
    "eval_rewards/accuracies": 0.5229970216751099,
    "eval_rewards/chosen": -1.786582350730896,
    "eval_rewards/margins": 0.08598694205284119,
    "eval_rewards/rejected": -1.8725693225860596,
    "eval_runtime": 60.635,
    "eval_samples": 1345,
    "eval_samples_per_second": 22.182,
    "eval_semantic_entropy": 0.6461161971092224,
    "eval_steps_per_second": 5.558
}