eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.0242695808410645,
    "eval_logits/rejected": -1.7967454195022583,
    "eval_logps/chosen": -255.23193359375,
    "eval_logps/rejected": -212.03785705566406,
    "eval_loss": 18.33973503112793,
    "eval_rewards/accuracies": 0.7200000286102295,
    "eval_rewards/chosen": 0.029216337949037552,
    "eval_rewards/margins": 0.12978971004486084,
    "eval_rewards/rejected": -0.10057336091995239,
    "eval_runtime": 239.208,
    "eval_samples": 2000,
    "eval_samples_per_second": 8.361,
    "eval_steps_per_second": 0.523
}