eval_results.json

{
    "epoch": 3.0,
    "eval_logits/chosen": -2.2653753757476807,
    "eval_logits/rejected": -2.2358808517456055,
    "eval_logps/chosen": -253.11325073242188,
    "eval_logps/rejected": -228.80319213867188,
    "eval_loss": 0.5631424188613892,
    "eval_rewards/accuracies": 0.8208333253860474,
    "eval_rewards/chosen": 1.1576802730560303,
    "eval_rewards/margins": 2.821425676345825,
    "eval_rewards/rejected": -1.6637452840805054,
    "eval_runtime": 124.889,
    "eval_samples_per_second": 15.214,
    "eval_steps_per_second": 0.24
}