eval_results.json

{
    "epoch": 1.0,
    "eval_logits/chosen": -1.684511661529541,
    "eval_logits/rejected": -1.6239439249038696,
    "eval_logps/chosen": -2806.67041015625,
    "eval_logps/rejected": -2491.757568359375,
    "eval_loss": 31.98944854736328,
    "eval_rewards/accuracies": 0.5234375,
    "eval_rewards/chosen": 0.024405580013990402,
    "eval_rewards/margins": 0.0056530386209487915,
    "eval_rewards/rejected": 0.01875254511833191,
    "eval_runtime": 110.5807,
    "eval_samples": 2000,
    "eval_samples_per_second": 18.086,
    "eval_steps_per_second": 0.289
}