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

{
    "epoch": 1.0,
    "total_flos": 0.0,
    "train_loss": 0.4133688477908864,
    "train_runtime": 1436.8705,
    "train_samples": 43598,
    "train_samples_per_second": 30.342,
    "train_steps_per_second": 0.237
}