eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -2.2437589168548584,
    "eval_logits/rejected": -2.2388815879821777,
    "eval_logps/chosen": -34.025550842285156,
    "eval_logps/rejected": -37.502403259277344,
    "eval_loss": 0.6950117945671082,
    "eval_rewards/accuracies": 0.49003320932388306,
    "eval_rewards/chosen": 0.0036009380128234625,
    "eval_rewards/margins": -0.0020864540711045265,
    "eval_rewards/rejected": 0.005687391385436058,
    "eval_runtime": 145.772,
    "eval_samples": 343,
    "eval_samples_per_second": 2.353,
    "eval_steps_per_second": 0.295
}