eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.107180118560791,
    "eval_logits/rejected": -2.047370672225952,
    "eval_logps/chosen": -264.3012390136719,
    "eval_logps/rejected": -214.9073944091797,
    "eval_loss": 0.500950038433075,
    "eval_rewards/accuracies": 0.8583333492279053,
    "eval_rewards/chosen": 1.1422334909439087,
    "eval_rewards/margins": 2.9307103157043457,
    "eval_rewards/rejected": -1.7884769439697266,
    "eval_runtime": 125.5521,
    "eval_samples_per_second": 15.133,
    "eval_steps_per_second": 0.239
}