eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -1.680073618888855,
    "eval_logits/rejected": -1.7620524168014526,
    "eval_logps/chosen": -349.61962890625,
    "eval_logps/rejected": -92.87176513671875,
    "eval_loss": 0.5131469964981079,
    "eval_rewards/accuracies": 1.0,
    "eval_rewards/chosen": 0.3873787820339203,
    "eval_rewards/margins": 0.40528222918510437,
    "eval_rewards/rejected": -0.017903532832860947,
    "eval_runtime": 67.3461,
    "eval_samples": 3077,
    "eval_samples_per_second": 45.689,
    "eval_steps_per_second": 0.728
}