models/validator-cond-1B-orpo-iter1-collab-paper/eval_results.json

{
    "epoch": 5.807339449541285,
    "eval_log_odds_chosen": 0.7261541485786438,
    "eval_log_odds_ratio": -0.4359833896160126,
    "eval_logits/chosen": 0.41933315992355347,
    "eval_logits/rejected": 0.3965945541858673,
    "eval_logps/chosen": -0.10455876588821411,
    "eval_logps/rejected": -0.19739307463169098,
    "eval_loss": 0.14815710484981537,
    "eval_nll_loss": 0.10455876588821411,
    "eval_rewards/accuracies": 0.8965517282485962,
    "eval_rewards/chosen": -0.010455875657498837,
    "eval_rewards/margins": 0.009283430874347687,
    "eval_rewards/rejected": -0.019739307463169098,
    "eval_runtime": 2.1352,
    "eval_samples": 29,
    "eval_samples_per_second": 13.582,
    "eval_steps_per_second": 13.582
}