eval_results.json

{
    "epoch": 4.0,
    "eval_logits/chosen": -2.2455787658691406,
    "eval_logits/rejected": -2.2406980991363525,
    "eval_logps/chosen": -33.94891357421875,
    "eval_logps/rejected": -37.46136474609375,
    "eval_loss": 0.7456009984016418,
    "eval_rewards/accuracies": 0.5365448594093323,
    "eval_rewards/chosen": 0.051383573561906815,
    "eval_rewards/margins": 0.018230266869068146,
    "eval_rewards/rejected": 0.03315330296754837,
    "eval_runtime": 145.63,
    "eval_samples": 343,
    "eval_samples_per_second": 2.355,
    "eval_steps_per_second": 0.295
}