llama-3-8b-base-margin-dpo-hh-harmless-8xh200/train_results.json

{
    "epoch": 1.0,
    "total_flos": 0.0,
    "train_loss": 0.5836806095007694,
    "train_runtime": 1387.0612,
    "train_samples": 42336,
    "train_samples_per_second": 30.522,
    "train_steps_per_second": 0.238
}