models/fixer-1B-orpo-iter1/eval_results.json

{
    "epoch": 14.285714285714286,
    "eval_log_odds_chosen": 3.4341936111450195,
    "eval_log_odds_ratio": -0.0597212128341198,
    "eval_logits/chosen": 0.3752058744430542,
    "eval_logits/rejected": 0.376191109418869,
    "eval_logps/chosen": -0.052630532532930374,
    "eval_logps/rejected": -0.6063037514686584,
    "eval_loss": 0.11235173791646957,
    "eval_nll_loss": 0.052630528807640076,
    "eval_rewards/accuracies": 1.0,
    "eval_rewards/chosen": -0.052630532532930374,
    "eval_rewards/margins": 0.5536733269691467,
    "eval_rewards/rejected": -0.6063037514686584,
    "eval_runtime": 0.775,
    "eval_samples": 12,
    "eval_samples_per_second": 15.483,
    "eval_steps_per_second": 15.483
}