llama-3-8b-base-epsilon-dpo-hh-harmless-4xh200-batch-64-20260418-003215/train_results.json

{
    "epoch": 0.999244142101285,
    "total_flos": 0.0,
    "train_loss": 1.1175190241903112,
    "train_runtime": 3196.4458,
    "train_samples": 42336,
    "train_samples_per_second": 13.245,
    "train_steps_per_second": 0.207
}