eval_results.json

{
    "epoch": 2.0,
    "eval_logits/chosen": -2.708155632019043,
    "eval_logits/rejected": -2.692856550216675,
    "eval_logps/chosen": -71.00025177001953,
    "eval_logps/rejected": -79.98316192626953,
    "eval_loss": 0.8595444560050964,
    "eval_rewards/accuracies": 0.3392857015132904,
    "eval_rewards/chosen": 1.850224494934082,
    "eval_rewards/margins": 1.243270993232727,
    "eval_rewards/rejected": 0.6069534420967102,
    "eval_runtime": 113.8695,
    "eval_samples": 2000,
    "eval_samples_per_second": 17.564,
    "eval_steps_per_second": 0.553
}